def _run(upconvnet_file_name, top_example_dir_name, first_spc_date_string,
last_spc_date_string, num_examples_per_date, downsampling_keys,
downsampling_values, top_output_dir_name):
"""Makes predictions from trained upconvnet.
This is effectively the main method.
:param upconvnet_file_name: See documentation at top of file.
:param top_example_dir_name: Same.
:param first_spc_date_string: Same.
:param last_spc_date_string: Same.
:param num_examples_per_date: Same.
:param downsampling_keys: Same.
:param downsampling_values: Same.
:param top_output_dir_name: Same.
"""
# Process input args.
print('Reading upconvnet from: "{0:s}"...'.format(upconvnet_file_name))
upconvnet_model_object = cnn.read_model(upconvnet_file_name)
upconvnet_metafile_name = cnn.find_metafile(upconvnet_file_name)
print('Reading upconvnet metadata from: "{0:s}"...'.format(
upconvnet_metafile_name))
upconvnet_metadata_dict = upconvnet.read_model_metadata(
upconvnet_metafile_name)
cnn_file_name = upconvnet_metadata_dict[upconvnet.CNN_FILE_KEY]
print('Reading CNN from: "{0:s}"...'.format(cnn_file_name))
cnn_model_object = cnn.read_model(cnn_file_name)
cnn_metafile_name = cnn.find_metafile(cnn_file_name)
print('Reading CNN metadata from: "{0:s}"...'.format(cnn_metafile_name))
cnn_metadata_dict = cnn.read_model_metadata(cnn_metafile_name)
training_option_dict = cnn_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
if len(downsampling_keys) > 1:
downsampling_dict = dict(
list(zip(downsampling_keys, downsampling_values)))
else:
downsampling_dict = None
training_option_dict[
trainval_io.SAMPLING_FRACTIONS_KEY] = downsampling_dict
training_option_dict[trainval_io.NUM_EXAMPLES_PER_BATCH_KEY] = (
NUM_EXAMPLES_PER_BATCH)
training_option_dict[
trainval_io.FIRST_STORM_TIME_KEY] = EARLY_TIME_UNIX_SEC
training_option_dict[trainval_io.LAST_STORM_TIME_KEY] = LATE_TIME_UNIX_SEC
# Find example files.
example_file_names = input_examples.find_many_example_files(
top_directory_name=top_example_dir_name,
shuffled=False,
first_spc_date_string=first_spc_date_string,
last_spc_date_string=last_spc_date_string,
raise_error_if_any_missing=False)
# Do dirty work.
for this_example_file_name in example_file_names:
_apply_upconvnet_one_file(
example_file_name=this_example_file_name,
num_examples=num_examples_per_date,
upconvnet_model_object=upconvnet_model_object,
cnn_model_object=cnn_model_object,
cnn_metadata_dict=cnn_metadata_dict,
cnn_feature_layer_name=upconvnet_metadata_dict[
upconvnet.CNN_FEATURE_LAYER_KEY],
upconvnet_file_name=upconvnet_file_name,
top_output_dir_name=top_output_dir_name)
print(SEPARATOR_STRING)
def _run(input_cnn_file_name, input_upconvnet_file_name,
cnn_feature_layer_name, top_training_dir_name,
first_training_time_string, last_training_time_string,
top_validation_dir_name, first_validation_time_string,
last_validation_time_string, num_examples_per_batch, num_epochs,
num_training_batches_per_epoch, num_validation_batches_per_epoch,
output_model_file_name):
"""Trains upconvnet.
This is effectively the main method.
:param input_cnn_file_name: See documentation at top of file.
:param input_upconvnet_file_name: Same.
:param cnn_feature_layer_name: Same.
:param top_training_dir_name: Same.
:param first_training_time_string: Same.
:param last_training_time_string: Same.
:param top_validation_dir_name: Same.
:param first_validation_time_string: Same.
:param last_validation_time_string: Same.
:param num_examples_per_batch: Same.
:param num_epochs: Same.
:param num_training_batches_per_epoch: Same.
:param num_validation_batches_per_epoch: Same.
:param output_model_file_name: Same.
"""
# Find training and validation files.
first_training_time_unix_sec = time_conversion.string_to_unix_sec(
first_training_time_string, TIME_FORMAT)
last_training_time_unix_sec = time_conversion.string_to_unix_sec(
last_training_time_string, TIME_FORMAT)
first_validation_time_unix_sec = time_conversion.string_to_unix_sec(
first_validation_time_string, TIME_FORMAT)
last_validation_time_unix_sec = time_conversion.string_to_unix_sec(
last_validation_time_string, TIME_FORMAT)
training_file_names = input_examples.find_many_example_files(
top_directory_name=top_training_dir_name,
shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER,
raise_error_if_any_missing=False)
validation_file_names = input_examples.find_many_example_files(
top_directory_name=top_validation_dir_name,
shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER,
raise_error_if_any_missing=False)
print 'Reading trained CNN from: "{0:s}"...'.format(input_cnn_file_name)
cnn_model_object = cnn.read_model(input_cnn_file_name)
cnn_model_object.summary()
print SEPARATOR_STRING
cnn_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(input_cnn_file_name)[0])
print 'Reading CNN metadata from: "{0:s}"...'.format(cnn_metafile_name)
cnn_metadata_dict = cnn.read_model_metadata(cnn_metafile_name)
print 'Reading upconvnet architecture from: "{0:s}"...'.format(
input_upconvnet_file_name)
upconvnet_model_object = cnn.read_model(input_upconvnet_file_name)
upconvnet_model_object = keras.models.clone_model(upconvnet_model_object)
# TODO(thunderhoser): This is a HACK.
upconvnet_model_object.compile(loss=keras.losses.mean_squared_error,
optimizer=keras.optimizers.Adam())
print SEPARATOR_STRING
upconvnet_model_object.summary()
print SEPARATOR_STRING
upconvnet_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(output_model_file_name)[0])
print 'Writing upconvnet metadata to: "{0:s}"...'.format(
upconvnet_metafile_name)
upconvnet.write_model_metadata(
cnn_file_name=input_cnn_file_name,
cnn_feature_layer_name=cnn_feature_layer_name,
num_epochs=num_epochs,
num_examples_per_batch=num_examples_per_batch,
num_training_batches_per_epoch=num_training_batches_per_epoch,
training_example_file_names=training_file_names,
first_training_time_unix_sec=first_training_time_unix_sec,
last_training_time_unix_sec=last_training_time_unix_sec,
num_validation_batches_per_epoch=num_validation_batches_per_epoch,
validation_example_file_names=validation_file_names,
first_validation_time_unix_sec=first_validation_time_unix_sec,
last_validation_time_unix_sec=last_validation_time_unix_sec,
pickle_file_name=upconvnet_metafile_name)
print SEPARATOR_STRING
upconvnet.train_upconvnet(
upconvnet_model_object=upconvnet_model_object,
output_model_file_name=output_model_file_name,
cnn_model_object=cnn_model_object,
cnn_feature_layer_name=cnn_feature_layer_name,
cnn_metadata_dict=cnn_metadata_dict,
num_epochs=num_epochs,
num_examples_per_batch=num_examples_per_batch,
num_training_batches_per_epoch=num_training_batches_per_epoch,
training_example_file_names=training_file_names,
first_training_time_unix_sec=first_training_time_unix_sec,
last_training_time_unix_sec=last_training_time_unix_sec,
num_validation_batches_per_epoch=num_validation_batches_per_epoch,
validation_example_file_names=validation_file_names,
first_validation_time_unix_sec=first_validation_time_unix_sec,
last_validation_time_unix_sec=last_validation_time_unix_sec)
def _run(input_cnn_file_name, input_upconvnet_file_name,
cnn_feature_layer_name, top_training_dir_name,
first_training_time_string, last_training_time_string,
top_validation_dir_name, first_validation_time_string,
last_validation_time_string, num_examples_per_batch, num_epochs,
num_training_batches_per_epoch, num_validation_batches_per_epoch,
output_dir_name):
"""Trains upconvnet.
This is effectively the main method.
:param input_cnn_file_name: See documentation at top of file.
:param input_upconvnet_file_name: Same.
:param cnn_feature_layer_name: Same.
:param top_training_dir_name: Same.
:param first_training_time_string: Same.
:param last_training_time_string: Same.
:param top_validation_dir_name: Same.
:param first_validation_time_string: Same.
:param last_validation_time_string: Same.
:param num_examples_per_batch: Same.
:param num_epochs: Same.
:param num_training_batches_per_epoch: Same.
:param num_validation_batches_per_epoch: Same.
:param output_dir_name: Same.
"""
file_system_utils.mkdir_recursive_if_necessary(
directory_name=output_dir_name)
# argument_file_name = '{0:s}/input_args.p'.format(output_dir_name)
# print('Writing input args to: "{0:s}"...'.format(argument_file_name))
#
# argument_file_handle = open(argument_file_name, 'wb')
# pickle.dump(INPUT_ARG_OBJECT.__dict__, argument_file_handle)
# argument_file_handle.close()
#
# return
# Process input args.
first_training_time_unix_sec = time_conversion.string_to_unix_sec(
first_training_time_string, TIME_FORMAT)
last_training_time_unix_sec = time_conversion.string_to_unix_sec(
last_training_time_string, TIME_FORMAT)
first_validation_time_unix_sec = time_conversion.string_to_unix_sec(
first_validation_time_string, TIME_FORMAT)
last_validation_time_unix_sec = time_conversion.string_to_unix_sec(
last_validation_time_string, TIME_FORMAT)
# Find training and validation files.
training_file_names = input_examples.find_many_example_files(
top_directory_name=top_training_dir_name,
shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER,
raise_error_if_any_missing=False)
validation_file_names = input_examples.find_many_example_files(
top_directory_name=top_validation_dir_name,
shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER,
raise_error_if_any_missing=False)
# Read trained CNN.
print('Reading trained CNN from: "{0:s}"...'.format(input_cnn_file_name))
cnn_model_object = cnn.read_model(input_cnn_file_name)
cnn_model_object.summary()
print(SEPARATOR_STRING)
cnn_metafile_name = cnn.find_metafile(model_file_name=input_cnn_file_name,
raise_error_if_missing=True)
print('Reading CNN metadata from: "{0:s}"...'.format(cnn_metafile_name))
cnn_metadata_dict = cnn.read_model_metadata(cnn_metafile_name)
# Read architecture.
print('Reading upconvnet architecture from: "{0:s}"...'.format(
input_upconvnet_file_name))
upconvnet_model_object = cnn.read_model(input_upconvnet_file_name)
# upconvnet_model_object = keras.models.clone_model(upconvnet_model_object)
# TODO(thunderhoser): This is a HACK.
upconvnet_model_object.compile(loss=keras.losses.mean_squared_error,
optimizer=keras.optimizers.Adam())
upconvnet_model_object.summary()
print(SEPARATOR_STRING)
upconvnet_metafile_name = cnn.find_metafile(
model_file_name='{0:s}/foo.h5'.format(output_dir_name),
raise_error_if_missing=False)
print('Writing upconvnet metadata to: "{0:s}"...'.format(
upconvnet_metafile_name))
upconvnet.write_model_metadata(
cnn_file_name=input_cnn_file_name,
cnn_feature_layer_name=cnn_feature_layer_name,
num_epochs=num_epochs,
num_examples_per_batch=num_examples_per_batch,
num_training_batches_per_epoch=num_training_batches_per_epoch,
training_example_file_names=training_file_names,
first_training_time_unix_sec=first_training_time_unix_sec,
last_training_time_unix_sec=last_training_time_unix_sec,
num_validation_batches_per_epoch=num_validation_batches_per_epoch,
validation_example_file_names=validation_file_names,
first_validation_time_unix_sec=first_validation_time_unix_sec,
last_validation_time_unix_sec=last_validation_time_unix_sec,
pickle_file_name=upconvnet_metafile_name)
print(SEPARATOR_STRING)
upconvnet.train_upconvnet(
upconvnet_model_object=upconvnet_model_object,
output_dir_name=output_dir_name,
cnn_model_object=cnn_model_object,
cnn_metadata_dict=cnn_metadata_dict,
cnn_feature_layer_name=cnn_feature_layer_name,
num_epochs=num_epochs,
num_examples_per_batch=num_examples_per_batch,
num_training_batches_per_epoch=num_training_batches_per_epoch,
training_example_file_names=training_file_names,
first_training_time_unix_sec=first_training_time_unix_sec,
last_training_time_unix_sec=last_training_time_unix_sec,
num_validation_batches_per_epoch=num_validation_batches_per_epoch,
validation_example_file_names=validation_file_names,
first_validation_time_unix_sec=first_validation_time_unix_sec,
last_validation_time_unix_sec=last_validation_time_unix_sec)
def _train_one_upconvnet(gpu_queue, argument_dict):
"""Trains one upconvolutional network.
:param gpu_queue: GPU queue (instance of `multiprocessing.Manager.Queue`).
:param argument_dict: Dictionary of CNN arguments, where each key is an
input arg to the script train_upconvnet.py.
"""
import keras
from keras import backend as K
import tensorflow
from gewittergefahr.deep_learning import cnn
from gewittergefahr.deep_learning import upconvnet
from gewittergefahr.scripts import train_upconvnet
gpu_index = -1
try:
# Deal with GPU business.
gpu_index = int(gpu_queue.get())
os.environ['CUDA_VISIBLE_DEVICES'] = '{0:d}'.format(gpu_index)
session_object = tensorflow.Session(
config=tensorflow.ConfigProto(
intra_op_parallelism_threads=7, inter_op_parallelism_threads=7,
allow_soft_placement=False, log_device_placement=False,
gpu_options=tensorflow.GPUOptions(allow_growth=True)
)
)
K.set_session(session_object)
# Write metadata.
upconvnet_metadata_dict = _write_metadata_one_model(argument_dict)
# Read trained CNN.
cnn_file_name = argument_dict[train_upconvnet.CNN_FILE_ARG_NAME]
cnn_metafile_name = cnn.find_metafile(cnn_file_name)
print('Reading CNN metadata from: "{0:s}"...'.format(cnn_metafile_name))
cnn_metadata_dict = cnn.read_model_metadata(cnn_metafile_name)
upconvnet_template_file_name = argument_dict[
train_upconvnet.UPCONVNET_FILE_ARG_NAME]
with tensorflow.device('/gpu:0'):
print('Reading trained CNN from: "{0:s}"...'.format(cnn_file_name))
cnn_model_object = cnn.read_model(cnn_file_name)
print('Reading upconvnet architecture from: "{0:s}"...'.format(
upconvnet_template_file_name
))
upconvnet_model_object = cnn.read_model(
upconvnet_template_file_name)
upconvnet_model_object.compile(
loss=keras.losses.mean_squared_error,
optimizer=keras.optimizers.Adam()
)
print(SEPARATOR_STRING)
upconvnet_model_object.summary()
print(SEPARATOR_STRING)
# Train upconvnet.
print('Training upconvnet on GPU {0:d}...'.format(gpu_index))
print(SEPARATOR_STRING)
upconvnet.train_upconvnet(
upconvnet_model_object=upconvnet_model_object,
output_dir_name=argument_dict[train_upconvnet.OUTPUT_DIR_ARG_NAME],
cnn_model_object=cnn_model_object,
cnn_metadata_dict=cnn_metadata_dict,
cnn_feature_layer_name=
upconvnet_metadata_dict[upconvnet.CNN_FEATURE_LAYER_KEY],
num_epochs=upconvnet_metadata_dict[upconvnet.NUM_EPOCHS_KEY],
num_examples_per_batch=
upconvnet_metadata_dict[upconvnet.NUM_EXAMPLES_PER_BATCH_KEY],
num_training_batches_per_epoch=
upconvnet_metadata_dict[upconvnet.NUM_TRAINING_BATCHES_KEY],
training_example_file_names=
upconvnet_metadata_dict[upconvnet.TRAINING_FILES_KEY],
first_training_time_unix_sec=
upconvnet_metadata_dict[upconvnet.FIRST_TRAINING_TIME_KEY],
last_training_time_unix_sec=
upconvnet_metadata_dict[upconvnet.LAST_TRAINING_TIME_KEY],
num_validation_batches_per_epoch=
upconvnet_metadata_dict[upconvnet.NUM_VALIDATION_BATCHES_KEY],
validation_example_file_names=
upconvnet_metadata_dict[upconvnet.VALIDATION_FILES_KEY],
first_validation_time_unix_sec=
upconvnet_metadata_dict[upconvnet.FIRST_VALIDATION_TIME_KEY],
last_validation_time_unix_sec=
upconvnet_metadata_dict[upconvnet.LAST_VALIDATION_TIME_KEY]
)
session_object.close()
del session_object
gpu_queue.put(gpu_index)
except Exception as this_exception:
if gpu_index >= 0:
gpu_queue.put(gpu_index)
print(traceback.format_exc())
raise this_exception
def _run(model_file_name, example_file_name, first_time_string,
last_time_string, top_output_dir_name):
"""Applies CNN to one example file.
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param example_file_name: Same.
:param first_time_string: Same.
:param last_time_string: Same.
:param top_output_dir_name: Same.
"""
print('Reading model from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
model_directory_name, _ = os.path.split(model_file_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(model_directory_name)
print('Reading metadata from: "{0:s}"...'.format(model_metafile_name))
model_metadata_dict = cnn.read_model_metadata(model_metafile_name)
training_option_dict = model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
first_time_unix_sec = time_conversion.string_to_unix_sec(
first_time_string, INPUT_TIME_FORMAT)
last_time_unix_sec = time_conversion.string_to_unix_sec(
last_time_string, INPUT_TIME_FORMAT)
training_option_dict[trainval_io.SAMPLING_FRACTIONS_KEY] = None
training_option_dict[trainval_io.EXAMPLE_FILES_KEY] = [example_file_name]
training_option_dict[
trainval_io.FIRST_STORM_TIME_KEY] = first_time_unix_sec
training_option_dict[trainval_io.LAST_STORM_TIME_KEY] = last_time_unix_sec
if model_metadata_dict[cnn.LAYER_OPERATIONS_KEY] is not None:
generator_object = testing_io.gridrad_generator_2d_reduced(
option_dict=training_option_dict,
list_of_operation_dicts=model_metadata_dict[
cnn.LAYER_OPERATIONS_KEY],
num_examples_total=LARGE_INTEGER)
elif model_metadata_dict[cnn.CONV_2D3D_KEY]:
generator_object = testing_io.myrorss_generator_2d3d(
option_dict=training_option_dict, num_examples_total=LARGE_INTEGER)
else:
generator_object = testing_io.generator_2d_or_3d(
option_dict=training_option_dict, num_examples_total=LARGE_INTEGER)
include_soundings = (training_option_dict[trainval_io.SOUNDING_FIELDS_KEY]
is not None)
try:
storm_object_dict = next(generator_object)
except StopIteration:
storm_object_dict = None
print(SEPARATOR_STRING)
if storm_object_dict is not None:
observed_labels = storm_object_dict[testing_io.TARGET_ARRAY_KEY]
list_of_predictor_matrices = storm_object_dict[
testing_io.INPUT_MATRICES_KEY]
if include_soundings:
sounding_matrix = list_of_predictor_matrices[-1]
else:
sounding_matrix = None
if model_metadata_dict[cnn.CONV_2D3D_KEY]:
if training_option_dict[trainval_io.UPSAMPLE_REFLECTIVITY_KEY]:
class_probability_matrix = cnn.apply_2d_or_3d_cnn(
model_object=model_object,
radar_image_matrix=list_of_predictor_matrices[0],
sounding_matrix=sounding_matrix,
verbose=True)
else:
class_probability_matrix = cnn.apply_2d3d_cnn(
model_object=model_object,
reflectivity_matrix_dbz=list_of_predictor_matrices[0],
azimuthal_shear_matrix_s01=list_of_predictor_matrices[1],
sounding_matrix=sounding_matrix,
verbose=True)
else:
class_probability_matrix = cnn.apply_2d_or_3d_cnn(
model_object=model_object,
radar_image_matrix=list_of_predictor_matrices[0],
sounding_matrix=sounding_matrix,
verbose=True)
print(SEPARATOR_STRING)
num_examples = class_probability_matrix.shape[0]
for k in [0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100]:
print(
'{0:d}th percentile of {1:d} forecast probs = {2:.4f}'.format(
k, num_examples,
numpy.percentile(class_probability_matrix[:, 1], k)))
print('\n')
target_param_dict = target_val_utils.target_name_to_params(
training_option_dict[trainval_io.TARGET_NAME_KEY])
event_type_string = target_param_dict[target_val_utils.EVENT_TYPE_KEY]
if event_type_string == linkage.TORNADO_EVENT_STRING:
genesis_only = False
elif event_type_string == linkage.TORNADOGENESIS_EVENT_STRING:
genesis_only = True
else:
genesis_only = None
target_name = target_val_utils.target_params_to_name(
min_lead_time_sec=target_param_dict[
target_val_utils.MIN_LEAD_TIME_KEY],
max_lead_time_sec=target_param_dict[
target_val_utils.MAX_LEAD_TIME_KEY],
min_link_distance_metres=target_param_dict[
target_val_utils.MIN_LINKAGE_DISTANCE_KEY],
max_link_distance_metres=10000.,
genesis_only=genesis_only)
output_file_name = prediction_io.find_file(
top_prediction_dir_name=top_output_dir_name,
first_init_time_unix_sec=first_time_unix_sec,
last_init_time_unix_sec=last_time_unix_sec,
gridded=False,
raise_error_if_missing=False)
print('Writing "{0:s}" predictions to: "{1:s}"...'.format(
target_name, output_file_name))
if storm_object_dict is None:
num_output_neurons = (
model_object.layers[-1].output.get_shape().as_list()[-1])
num_classes = max([num_output_neurons, 2])
class_probability_matrix = numpy.full((0, num_classes), numpy.nan)
prediction_io.write_ungridded_predictions(
netcdf_file_name=output_file_name,
class_probability_matrix=class_probability_matrix,
storm_ids=[],
storm_times_unix_sec=numpy.array([], dtype=int),
target_name=target_name,
observed_labels=numpy.array([], dtype=int))
return
prediction_io.write_ungridded_predictions(
netcdf_file_name=output_file_name,
class_probability_matrix=class_probability_matrix,
storm_ids=storm_object_dict[testing_io.FULL_IDS_KEY],
storm_times_unix_sec=storm_object_dict[testing_io.STORM_TIMES_KEY],
target_name=target_name,
observed_labels=observed_labels)
def _run(model_file_name, top_example_dir_name, first_spc_date_string,
last_spc_date_string, num_examples, do_backwards_test,
separate_radar_heights, downsampling_keys, downsampling_values,
num_bootstrap_reps, output_file_name):
"""Runs permutation test for predictor importance.
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param top_example_dir_name: Same.
:param first_spc_date_string: Same.
:param last_spc_date_string: Same.
:param num_examples: Same.
:param do_backwards_test: Same.
:param separate_radar_heights: Same.
:param downsampling_keys: Same.
:param downsampling_values: Same.
:param num_bootstrap_reps: Same.
:param output_file_name: Same.
"""
print('Reading model from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
metafile_name = cnn.find_metafile(model_file_name=model_file_name)
print('Reading metadata from: "{0:s}"...'.format(metafile_name))
cnn_metadata_dict = cnn.read_model_metadata(metafile_name)
if len(downsampling_keys) > 1:
downsampling_dict = dict(
list(zip(downsampling_keys, downsampling_values)))
else:
downsampling_dict = None
training_option_dict = cnn_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
training_option_dict[
trainval_io.SAMPLING_FRACTIONS_KEY] = downsampling_dict
example_file_names = input_examples.find_many_example_files(
top_directory_name=top_example_dir_name,
shuffled=False,
first_spc_date_string=first_spc_date_string,
last_spc_date_string=last_spc_date_string,
raise_error_if_any_missing=False)
training_option_dict[trainval_io.EXAMPLE_FILES_KEY] = example_file_names
training_option_dict[trainval_io.FIRST_STORM_TIME_KEY] = (
time_conversion.get_start_of_spc_date(first_spc_date_string))
training_option_dict[trainval_io.LAST_STORM_TIME_KEY] = (
time_conversion.get_end_of_spc_date(last_spc_date_string))
training_option_dict[trainval_io.NUM_EXAMPLES_PER_BATCH_KEY] = (
NUM_EXAMPLES_PER_BATCH)
if cnn_metadata_dict[cnn.LAYER_OPERATIONS_KEY] is not None:
generator_object = testing_io.gridrad_generator_2d_reduced(
option_dict=training_option_dict,
desired_num_examples=num_examples,
list_of_operation_dicts=cnn_metadata_dict[
cnn.LAYER_OPERATIONS_KEY])
elif cnn_metadata_dict[cnn.CONV_2D3D_KEY]:
generator_object = testing_io.myrorss_generator_2d3d(
option_dict=training_option_dict,
desired_num_examples=num_examples)
else:
generator_object = testing_io.generator_2d_or_3d(
option_dict=training_option_dict,
desired_num_examples=num_examples)
full_storm_id_strings = []
storm_times_unix_sec = numpy.array([], dtype=int)
target_values = numpy.array([], dtype=int)
predictor_matrices = None
print(SEPARATOR_STRING)
for _ in range(len(example_file_names)):
try:
this_storm_object_dict = next(generator_object)
print(SEPARATOR_STRING)
except StopIteration:
break
full_storm_id_strings += this_storm_object_dict[
testing_io.FULL_IDS_KEY]
storm_times_unix_sec = numpy.concatenate(
(storm_times_unix_sec,
this_storm_object_dict[testing_io.STORM_TIMES_KEY]))
these_target_values = this_storm_object_dict[
testing_io.TARGET_ARRAY_KEY]
if len(these_target_values.shape) > 1:
these_target_values = numpy.argmax(these_target_values, axis=1)
target_values = numpy.concatenate((target_values, these_target_values))
these_predictor_matrices = this_storm_object_dict[
testing_io.INPUT_MATRICES_KEY]
if predictor_matrices is None:
predictor_matrices = copy.deepcopy(these_predictor_matrices)
else:
for k in range(len(predictor_matrices)):
predictor_matrices[k] = numpy.concatenate(
(predictor_matrices[k], these_predictor_matrices[k]))
print(SEPARATOR_STRING)
correlation_matrix, predictor_names = correlation.get_pearson_correlations(
predictor_matrices=predictor_matrices,
cnn_metadata_dict=cnn_metadata_dict,
separate_radar_heights=separate_radar_heights)
print(SEPARATOR_STRING)
num_predictors = len(predictor_names)
for i in range(num_predictors):
for j in range(i, num_predictors):
print(('Pearson correlation between "{0:s}" and "{1:s}" = {2:.3f}'
).format(predictor_names[i], predictor_names[j],
correlation_matrix[i, j]))
print(SEPARATOR_STRING)
if do_backwards_test:
result_dict = permutation.run_backwards_test(
model_object=model_object,
predictor_matrices=predictor_matrices,
target_values=target_values,
cnn_metadata_dict=cnn_metadata_dict,
cost_function=permutation_utils.negative_auc_function,
separate_radar_heights=separate_radar_heights,
num_bootstrap_reps=num_bootstrap_reps)
else:
result_dict = permutation.run_forward_test(
model_object=model_object,
predictor_matrices=predictor_matrices,
target_values=target_values,
cnn_metadata_dict=cnn_metadata_dict,
cost_function=permutation_utils.negative_auc_function,
separate_radar_heights=separate_radar_heights,
num_bootstrap_reps=num_bootstrap_reps)
print(SEPARATOR_STRING)
result_dict[permutation_utils.MODEL_FILE_KEY] = model_file_name
result_dict[permutation_utils.TARGET_VALUES_KEY] = target_values
result_dict[permutation_utils.FULL_IDS_KEY] = full_storm_id_strings
result_dict[permutation_utils.STORM_TIMES_KEY] = storm_times_unix_sec
print('Writing results to: "{0:s}"...'.format(output_file_name))
permutation_utils.write_results(result_dict=result_dict,
pickle_file_name=output_file_name)
def _run(model_file_name, top_example_dir_name, storm_metafile_name,
num_examples, do_backwards_test, separate_radar_heights,
num_bootstrap_reps, output_file_name):
"""Runs permutation test with specific examples (storm objects).
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param top_example_dir_name: Same.
:param storm_metafile_name: Same.
:param num_examples: Same.
:param do_backwards_test: Same.
:param separate_radar_heights: Same.
:param num_bootstrap_reps: Same.
:param output_file_name: Same.
"""
print('Reading model from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
metafile_name = cnn.find_metafile(model_file_name=model_file_name)
print('Reading metadata from: "{0:s}"...'.format(metafile_name))
cnn_metadata_dict = cnn.read_model_metadata(metafile_name)
training_option_dict = cnn_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
print(
'Reading storm metadata from: "{0:s}"...'.format(storm_metafile_name))
full_storm_id_strings, storm_times_unix_sec = (
tracking_io.read_ids_and_times(storm_metafile_name))
print(SEPARATOR_STRING)
if 0 < num_examples < len(full_storm_id_strings):
numpy.random.seed(RANDOM_SEED)
good_indices = numpy.random.permutation(len(full_storm_id_strings))
good_indices = good_indices[:num_examples]
full_storm_id_strings = [
full_storm_id_strings[k] for k in good_indices
]
storm_times_unix_sec = storm_times_unix_sec[good_indices]
example_dict = testing_io.read_predictors_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_full_id_strings=full_storm_id_strings,
desired_times_unix_sec=storm_times_unix_sec,
option_dict=training_option_dict,
layer_operation_dicts=cnn_metadata_dict[cnn.LAYER_OPERATIONS_KEY])
print(SEPARATOR_STRING)
predictor_matrices = example_dict[testing_io.INPUT_MATRICES_KEY]
target_values = example_dict[testing_io.TARGET_ARRAY_KEY]
correlation_matrix, predictor_names = correlation.get_pearson_correlations(
predictor_matrices=predictor_matrices,
cnn_metadata_dict=cnn_metadata_dict,
separate_radar_heights=separate_radar_heights)
print(SEPARATOR_STRING)
num_predictors = len(predictor_names)
for i in range(num_predictors):
for j in range(i, num_predictors):
print(('Pearson correlation between "{0:s}" and "{1:s}" = {2:.3f}'
).format(predictor_names[i], predictor_names[j],
correlation_matrix[i, j]))
print(SEPARATOR_STRING)
if do_backwards_test:
result_dict = permutation.run_backwards_test(
model_object=model_object,
predictor_matrices=predictor_matrices,
target_values=target_values,
cnn_metadata_dict=cnn_metadata_dict,
cost_function=permutation_utils.negative_auc_function,
separate_radar_heights=separate_radar_heights,
num_bootstrap_reps=num_bootstrap_reps)
else:
result_dict = permutation.run_forward_test(
model_object=model_object,
predictor_matrices=predictor_matrices,
target_values=target_values,
cnn_metadata_dict=cnn_metadata_dict,
cost_function=permutation_utils.negative_auc_function,
separate_radar_heights=separate_radar_heights,
num_bootstrap_reps=num_bootstrap_reps)
print(SEPARATOR_STRING)
result_dict[permutation_utils.MODEL_FILE_KEY] = model_file_name
result_dict[permutation_utils.TARGET_VALUES_KEY] = target_values
result_dict[permutation_utils.FULL_IDS_KEY] = full_storm_id_strings
result_dict[permutation_utils.STORM_TIMES_KEY] = storm_times_unix_sec
print('Writing results to: "{0:s}"...'.format(output_file_name))
permutation_utils.write_results(result_dict=result_dict,
pickle_file_name=output_file_name)
def _run(model_file_name, top_example_dir_name, first_spc_date_string,
last_spc_date_string, num_examples, num_bootstrap_reps,
confidence_level, class_fraction_keys, class_fraction_values,
output_dir_name):
"""Evaluates CNN (convolutional neural net) predictions.
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param top_example_dir_name: Same.
:param first_spc_date_string: Same.
:param last_spc_date_string: Same.
:param num_examples: Same.
:param num_bootstrap_reps: Same.
:param confidence_level: Same.
:param class_fraction_keys: Same.
:param class_fraction_values: Same.
:param output_dir_name: Same.
:raises: ValueError: if the model does multi-class classification.
"""
print('Reading model from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
num_output_neurons = (
model_object.layers[-1].output.get_shape().as_list()[-1])
if num_output_neurons > 2:
error_string = (
'The model has {0:d} output neurons, which suggests {0:d}-class '
'classification. This script handles only binary classification.'
).format(num_output_neurons)
raise ValueError(error_string)
soundings_only = False
if isinstance(model_object.input, list):
list_of_input_tensors = model_object.input
else:
list_of_input_tensors = [model_object.input]
if len(list_of_input_tensors) == 1:
these_spatial_dim = numpy.array(
list_of_input_tensors[0].get_shape().as_list()[1:-1], dtype=int)
soundings_only = len(these_spatial_dim) == 1
model_directory_name, _ = os.path.split(model_file_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(model_directory_name)
print('Reading metadata from: "{0:s}"...'.format(model_metafile_name))
model_metadata_dict = cnn.read_model_metadata(model_metafile_name)
training_option_dict = model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
if len(class_fraction_keys) > 1:
class_to_sampling_fraction_dict = dict(
list(zip(class_fraction_keys, class_fraction_values)))
else:
class_to_sampling_fraction_dict = None
training_option_dict[
trainval_io.SAMPLING_FRACTIONS_KEY] = class_to_sampling_fraction_dict
example_file_names = input_examples.find_many_example_files(
top_directory_name=top_example_dir_name,
shuffled=False,
first_spc_date_string=first_spc_date_string,
last_spc_date_string=last_spc_date_string,
raise_error_if_any_missing=False)
training_option_dict[trainval_io.EXAMPLE_FILES_KEY] = example_file_names
training_option_dict[trainval_io.FIRST_STORM_TIME_KEY] = (
time_conversion.get_start_of_spc_date(first_spc_date_string))
training_option_dict[trainval_io.LAST_STORM_TIME_KEY] = (
time_conversion.get_end_of_spc_date(last_spc_date_string))
if soundings_only:
generator_object = testing_io.sounding_generator(
option_dict=training_option_dict, num_examples_total=num_examples)
elif model_metadata_dict[cnn.LAYER_OPERATIONS_KEY] is not None:
generator_object = testing_io.gridrad_generator_2d_reduced(
option_dict=training_option_dict,
list_of_operation_dicts=model_metadata_dict[
cnn.LAYER_OPERATIONS_KEY],
num_examples_total=num_examples)
elif model_metadata_dict[cnn.CONV_2D3D_KEY]:
generator_object = testing_io.myrorss_generator_2d3d(
option_dict=training_option_dict, num_examples_total=num_examples)
else:
generator_object = testing_io.generator_2d_or_3d(
option_dict=training_option_dict, num_examples_total=num_examples)
include_soundings = (training_option_dict[trainval_io.SOUNDING_FIELDS_KEY]
is not None)
forecast_probabilities = numpy.array([])
observed_labels = numpy.array([], dtype=int)
for _ in range(len(example_file_names)):
try:
this_storm_object_dict = next(generator_object)
print(SEPARATOR_STRING)
except StopIteration:
break
observed_labels = numpy.concatenate(
(observed_labels,
this_storm_object_dict[testing_io.TARGET_ARRAY_KEY]))
if soundings_only:
these_predictor_matrices = [
this_storm_object_dict[testing_io.SOUNDING_MATRIX_KEY]
]
else:
these_predictor_matrices = this_storm_object_dict[
testing_io.INPUT_MATRICES_KEY]
if include_soundings:
this_sounding_matrix = these_predictor_matrices[-1]
else:
this_sounding_matrix = None
if soundings_only:
this_probability_matrix = cnn.apply_cnn_soundings_only(
model_object=model_object,
sounding_matrix=this_sounding_matrix,
verbose=True)
elif model_metadata_dict[cnn.CONV_2D3D_KEY]:
if training_option_dict[trainval_io.UPSAMPLE_REFLECTIVITY_KEY]:
this_probability_matrix = cnn.apply_2d_or_3d_cnn(
model_object=model_object,
radar_image_matrix=these_predictor_matrices[0],
sounding_matrix=this_sounding_matrix,
verbose=True)
else:
this_probability_matrix = cnn.apply_2d3d_cnn(
model_object=model_object,
reflectivity_matrix_dbz=these_predictor_matrices[0],
azimuthal_shear_matrix_s01=these_predictor_matrices[1],
sounding_matrix=this_sounding_matrix,
verbose=True)
else:
this_probability_matrix = cnn.apply_2d_or_3d_cnn(
model_object=model_object,
radar_image_matrix=these_predictor_matrices[0],
sounding_matrix=this_sounding_matrix,
verbose=True)
print(SEPARATOR_STRING)
forecast_probabilities = numpy.concatenate(
(forecast_probabilities, this_probability_matrix[:, -1]))
model_eval_helper.run_evaluation(
forecast_probabilities=forecast_probabilities,
observed_labels=observed_labels,
num_bootstrap_reps=num_bootstrap_reps,
confidence_level=confidence_level,
output_dir_name=output_dir_name)
def _run(model_file_name, target_class, target_layer_name,
top_example_dir_name, storm_metafile_name, num_examples,
output_file_name):
"""Runs Grad-CAM (gradient-weighted class-activation maps).
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param target_class: Same.
:param target_layer_name: Same.
:param top_example_dir_name: Same.
:param storm_metafile_name: Same.
:param num_examples: Same.
:param output_file_name: Same.
"""
file_system_utils.mkdir_recursive_if_necessary(file_name=output_file_name)
# Read model and metadata.
print('Reading model from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(model_file_name)[0])
print(
'Reading model metadata from: "{0:s}"...'.format(model_metafile_name))
model_metadata_dict = cnn.read_model_metadata(model_metafile_name)
training_option_dict = model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
training_option_dict[trainval_io.REFLECTIVITY_MASK_KEY] = None
print(
'Reading storm metadata from: "{0:s}"...'.format(storm_metafile_name))
full_id_strings, storm_times_unix_sec = tracking_io.read_ids_and_times(
storm_metafile_name)
print(SEPARATOR_STRING)
if 0 < num_examples < len(full_id_strings):
full_id_strings = full_id_strings[:num_examples]
storm_times_unix_sec = storm_times_unix_sec[:num_examples]
list_of_input_matrices, sounding_pressure_matrix_pascals = (
testing_io.read_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_full_id_strings=full_id_strings,
desired_times_unix_sec=storm_times_unix_sec,
option_dict=training_option_dict,
list_of_layer_operation_dicts=model_metadata_dict[
cnn.LAYER_OPERATIONS_KEY]))
print(SEPARATOR_STRING)
list_of_cam_matrices = None
list_of_guided_cam_matrices = None
new_model_object = None
num_examples = len(full_id_strings)
for i in range(num_examples):
print('Running Grad-CAM for example {0:d} of {1:d}...'.format(
i + 1, num_examples))
these_input_matrices = [a[[i], ...] for a in list_of_input_matrices]
these_cam_matrices = gradcam.run_gradcam(
model_object=model_object,
list_of_input_matrices=these_input_matrices,
target_class=target_class,
target_layer_name=target_layer_name)
print('Running guided Grad-CAM for example {0:d} of {1:d}...'.format(
i + 1, num_examples))
these_guided_cam_matrices, new_model_object = (
gradcam.run_guided_gradcam(
orig_model_object=model_object,
list_of_input_matrices=these_input_matrices,
target_layer_name=target_layer_name,
list_of_cam_matrices=these_cam_matrices,
new_model_object=new_model_object))
if list_of_cam_matrices is None:
list_of_cam_matrices = copy.deepcopy(these_cam_matrices)
list_of_guided_cam_matrices = copy.deepcopy(
these_guided_cam_matrices)
else:
for j in range(len(these_cam_matrices)):
if list_of_cam_matrices[j] is None:
continue
list_of_cam_matrices[j] = numpy.concatenate(
(list_of_cam_matrices[j], these_cam_matrices[j]), axis=0)
list_of_guided_cam_matrices[j] = numpy.concatenate(
(list_of_guided_cam_matrices[j],
these_guided_cam_matrices[j]),
axis=0)
print(SEPARATOR_STRING)
upsample_refl = training_option_dict[trainval_io.UPSAMPLE_REFLECTIVITY_KEY]
if upsample_refl:
list_of_cam_matrices[0] = numpy.expand_dims(list_of_cam_matrices[0],
axis=-1)
num_channels = list_of_input_matrices[0].shape[-1]
list_of_cam_matrices[0] = numpy.repeat(a=list_of_cam_matrices[0],
repeats=num_channels,
axis=-1)
list_of_cam_matrices = trainval_io.separate_shear_and_reflectivity(
list_of_input_matrices=list_of_cam_matrices,
training_option_dict=training_option_dict)
list_of_cam_matrices[0] = list_of_cam_matrices[0][..., 0]
list_of_cam_matrices[1] = list_of_cam_matrices[1][..., 0]
list_of_guided_cam_matrices = trainval_io.separate_shear_and_reflectivity(
list_of_input_matrices=list_of_guided_cam_matrices,
training_option_dict=training_option_dict)
print('Denormalizing predictors...')
list_of_input_matrices = trainval_io.separate_shear_and_reflectivity(
list_of_input_matrices=list_of_input_matrices,
training_option_dict=training_option_dict)
list_of_input_matrices = model_interpretation.denormalize_data(
list_of_input_matrices=list_of_input_matrices,
model_metadata_dict=model_metadata_dict)
print('Writing class-activation maps to file: "{0:s}"...'.format(
output_file_name))
gradcam.write_standard_file(
pickle_file_name=output_file_name,
list_of_input_matrices=list_of_input_matrices,
list_of_cam_matrices=list_of_cam_matrices,
list_of_guided_cam_matrices=list_of_guided_cam_matrices,
model_file_name=model_file_name,
full_id_strings=full_id_strings,
storm_times_unix_sec=storm_times_unix_sec,
target_class=target_class,
target_layer_name=target_layer_name,
sounding_pressure_matrix_pascals=sounding_pressure_matrix_pascals)
def _run(model_file_name, target_class, target_layer_name,
top_example_dir_name, storm_metafile_name, num_examples,
output_file_name):
"""Runs Grad-CAM (gradient-weighted class-activation maps).
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param target_class: Same.
:param target_layer_name: Same.
:param top_example_dir_name: Same.
:param storm_metafile_name: Same.
:param num_examples: Same.
:param output_file_name: Same.
"""
file_system_utils.mkdir_recursive_if_necessary(file_name=output_file_name)
# Read model and metadata.
print 'Reading model from: "{0:s}"...'.format(model_file_name)
model_object = cnn.read_model(model_file_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(model_file_name)[0])
print 'Reading model metadata from: "{0:s}"...'.format(model_metafile_name)
model_metadata_dict = cnn.read_model_metadata(model_metafile_name)
training_option_dict = model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
training_option_dict[trainval_io.REFLECTIVITY_MASK_KEY] = None
print 'Reading storm metadata from: "{0:s}"...'.format(storm_metafile_name)
storm_ids, storm_times_unix_sec = tracking_io.read_ids_and_times(
storm_metafile_name)
print SEPARATOR_STRING
if 0 < num_examples < len(storm_ids):
storm_ids = storm_ids[:num_examples]
storm_times_unix_sec = storm_times_unix_sec[:num_examples]
list_of_input_matrices, sounding_pressure_matrix_pascals = (
testing_io.read_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_storm_ids=storm_ids,
desired_times_unix_sec=storm_times_unix_sec,
option_dict=training_option_dict,
list_of_layer_operation_dicts=model_metadata_dict[
cnn.LAYER_OPERATIONS_KEY]))
print SEPARATOR_STRING
class_activation_matrix = None
ggradcam_output_matrix = None
new_model_object = None
num_examples = len(storm_ids)
for i in range(num_examples):
print 'Running Grad-CAM for example {0:d} of {1:d}...'.format(
i + 1, num_examples)
these_input_matrices = [a[[i], ...] for a in list_of_input_matrices]
this_class_activation_matrix = gradcam.run_gradcam(
model_object=model_object,
list_of_input_matrices=these_input_matrices,
target_class=target_class,
target_layer_name=target_layer_name)
print 'Running guided Grad-CAM for example {0:d} of {1:d}...'.format(
i + 1, num_examples)
this_ggradcam_output_matrix, new_model_object = (
gradcam.run_guided_gradcam(
orig_model_object=model_object,
list_of_input_matrices=these_input_matrices,
target_layer_name=target_layer_name,
class_activation_matrix=this_class_activation_matrix,
new_model_object=new_model_object))
this_class_activation_matrix = numpy.expand_dims(
this_class_activation_matrix, axis=0)
this_ggradcam_output_matrix = numpy.expand_dims(
this_ggradcam_output_matrix, axis=0)
if class_activation_matrix is None:
class_activation_matrix = this_class_activation_matrix + 0.
ggradcam_output_matrix = this_ggradcam_output_matrix + 0.
else:
class_activation_matrix = numpy.concatenate(
(class_activation_matrix, this_class_activation_matrix),
axis=0)
ggradcam_output_matrix = numpy.concatenate(
(ggradcam_output_matrix, this_ggradcam_output_matrix), axis=0)
print SEPARATOR_STRING
print 'Denormalizing predictors...'
list_of_input_matrices = model_interpretation.denormalize_data(
list_of_input_matrices=list_of_input_matrices,
model_metadata_dict=model_metadata_dict)
print 'Writing class-activation maps to file: "{0:s}"...'.format(
output_file_name)
gradcam.write_standard_file(
pickle_file_name=output_file_name,
list_of_input_matrices=list_of_input_matrices,
class_activation_matrix=class_activation_matrix,
ggradcam_output_matrix=ggradcam_output_matrix,
model_file_name=model_file_name,
storm_ids=storm_ids,
storm_times_unix_sec=storm_times_unix_sec,
target_class=target_class,
target_layer_name=target_layer_name,
sounding_pressure_matrix_pascals=sounding_pressure_matrix_pascals)
def _run(model_file_name, init_function_name, storm_metafile_name,
num_examples, top_example_dir_name, component_type_string,
target_class, layer_name, neuron_indices, channel_index,
num_iterations, ideal_activation, learning_rate, output_file_name):
"""Runs backwards optimization on a trained CNN.
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param init_function_name: Same.
:param storm_metafile_name: Same.
:param num_examples: Same.
:param top_example_dir_name: Same.
:param component_type_string: Same.
:param target_class: Same.
:param layer_name: Same.
:param neuron_indices: Same.
:param channel_index: Same.
:param num_iterations: Same.
:param ideal_activation: Same.
:param learning_rate: Same.
:param output_file_name: Same.
"""
model_interpretation.check_component_type(component_type_string)
if ideal_activation <= 0:
ideal_activation = None
if init_function_name in ['', 'None']:
init_function_name = None
model_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(model_file_name)[0])
print 'Reading model metadata from: "{0:s}"...'.format(model_metafile_name)
model_metadata_dict = cnn.read_model_metadata(model_metafile_name)
if init_function_name is None:
print 'Reading storm metadata from: "{0:s}"...'.format(
storm_metafile_name)
storm_ids, storm_times_unix_sec = tracking_io.read_ids_and_times(
storm_metafile_name)
if 0 < num_examples < len(storm_ids):
storm_ids = storm_ids[:num_examples]
storm_times_unix_sec = storm_times_unix_sec[:num_examples]
list_of_init_matrices = testing_io.read_specific_examples(
desired_storm_ids=storm_ids,
desired_times_unix_sec=storm_times_unix_sec,
option_dict=model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY],
top_example_dir_name=top_example_dir_name,
list_of_layer_operation_dicts=model_metadata_dict[
cnn.LAYER_OPERATIONS_KEY])[0]
num_examples = list_of_init_matrices[0].shape[0]
print SEPARATOR_STRING
else:
storm_ids = None
storm_times_unix_sec = None
num_examples = 1
init_function = _create_initializer(
init_function_name=init_function_name,
model_metadata_dict=model_metadata_dict)
print 'Reading model from: "{0:s}"...'.format(model_file_name)
model_object = cnn.read_model(model_file_name)
list_of_optimized_matrices = None
for i in range(num_examples):
if init_function_name is None:
this_init_arg = [a[[i], ...] for a in list_of_init_matrices]
else:
this_init_arg = init_function
if component_type_string == CLASS_COMPONENT_TYPE_STRING:
print(
'\nOptimizing {0:d}th of {1:d} images for target class {2:d}...'
).format(i + 1, num_examples, target_class)
these_optimized_matrices = backwards_opt.optimize_input_for_class(
model_object=model_object,
target_class=target_class,
init_function_or_matrices=this_init_arg,
num_iterations=num_iterations,
learning_rate=learning_rate)
elif component_type_string == NEURON_COMPONENT_TYPE_STRING:
print(
'\nOptimizing {0:d}th of {1:d} images for neuron {2:s} in layer'
' "{3:s}"...').format(i + 1, num_examples, str(neuron_indices),
layer_name)
these_optimized_matrices = backwards_opt.optimize_input_for_neuron(
model_object=model_object,
layer_name=layer_name,
neuron_indices=neuron_indices,
init_function_or_matrices=this_init_arg,
num_iterations=num_iterations,
learning_rate=learning_rate,
ideal_activation=ideal_activation)
else:
print(
'\nOptimizing {0:d}th of {1:d} images for channel {2:d} in '
'layer "{3:s}"...').format(i + 1, num_examples, channel_index,
layer_name)
these_optimized_matrices = backwards_opt.optimize_input_for_channel(
model_object=model_object,
layer_name=layer_name,
channel_index=channel_index,
init_function_or_matrices=this_init_arg,
stat_function_for_neuron_activations=K.max,
num_iterations=num_iterations,
learning_rate=learning_rate,
ideal_activation=ideal_activation)
if list_of_optimized_matrices is None:
num_matrices = len(these_optimized_matrices)
list_of_optimized_matrices = [None] * num_matrices
for k in range(len(list_of_optimized_matrices)):
if list_of_optimized_matrices[k] is None:
list_of_optimized_matrices[
k] = these_optimized_matrices[k] + 0.
else:
list_of_optimized_matrices[k] = numpy.concatenate(
(list_of_optimized_matrices[k],
these_optimized_matrices[k]),
axis=0)
print SEPARATOR_STRING
print 'Denormalizing optimized examples...'
list_of_optimized_matrices = model_interpretation.denormalize_data(
list_of_input_matrices=list_of_optimized_matrices,
model_metadata_dict=model_metadata_dict)
if init_function_name is None:
print 'Denormalizing input examples...'
list_of_init_matrices = model_interpretation.denormalize_data(
list_of_input_matrices=list_of_init_matrices,
model_metadata_dict=model_metadata_dict)
this_init_arg = list_of_init_matrices
else:
this_init_arg = init_function_name + ''
print 'Writing results to: "{0:s}"...'.format(output_file_name)
backwards_opt.write_standard_file(
pickle_file_name=output_file_name,
list_of_optimized_matrices=list_of_optimized_matrices,
model_file_name=model_file_name,
init_function_name_or_matrices=this_init_arg,
num_iterations=num_iterations,
learning_rate=learning_rate,
component_type_string=component_type_string,
target_class=target_class,
layer_name=layer_name,
neuron_indices=neuron_indices,
channel_index=channel_index,
ideal_activation=ideal_activation,
storm_ids=storm_ids,
storm_times_unix_sec=storm_times_unix_sec)
def _run(cnn_file_name, upconvnet_file_name, top_example_dir_name,
baseline_storm_metafile_name, trial_storm_metafile_name,
num_baseline_examples, num_trial_examples, num_novel_examples,
cnn_feature_layer_name, percent_variance_to_keep, output_file_name):
"""Runs novelty detection.
This is effectively the main method.
:param cnn_file_name: See documentation at top of file.
:param upconvnet_file_name: Same.
:param top_example_dir_name: Same.
:param baseline_storm_metafile_name: Same.
:param trial_storm_metafile_name: Same.
:param num_baseline_examples: Same.
:param num_trial_examples: Same.
:param num_novel_examples: Same.
:param cnn_feature_layer_name: Same.
:param percent_variance_to_keep: Same.
:param output_file_name: Same.
:raises: ValueError: if dimensions of first CNN input matrix != dimensions
of upconvnet output.
"""
print('Reading trained CNN from: "{0:s}"...'.format(cnn_file_name))
cnn_model_object = cnn.read_model(cnn_file_name)
print('Reading trained upconvnet from: "{0:s}"...'.format(
upconvnet_file_name))
upconvnet_model_object = cnn.read_model(upconvnet_file_name)
_check_dimensions(cnn_model_object=cnn_model_object,
upconvnet_model_object=upconvnet_model_object)
print('Reading metadata for baseline examples from: "{0:s}"...'.format(
baseline_storm_metafile_name))
baseline_full_id_strings, baseline_times_unix_sec = (
tracking_io.read_ids_and_times(baseline_storm_metafile_name))
print('Reading metadata for trial examples from: "{0:s}"...'.format(
trial_storm_metafile_name))
trial_full_id_strings, trial_times_unix_sec = (
tracking_io.read_ids_and_times(trial_storm_metafile_name))
this_dict = _filter_examples(
trial_full_id_strings=trial_full_id_strings,
trial_times_unix_sec=trial_times_unix_sec,
num_trial_examples=num_trial_examples,
baseline_full_id_strings=baseline_full_id_strings,
baseline_times_unix_sec=baseline_times_unix_sec,
num_baseline_examples=num_baseline_examples,
num_novel_examples=num_novel_examples)
trial_full_id_strings = this_dict[TRIAL_STORM_IDS_KEY]
trial_times_unix_sec = this_dict[TRIAL_STORM_TIMES_KEY]
baseline_full_id_strings = this_dict[BASELINE_STORM_IDS_KEY]
baseline_times_unix_sec = this_dict[BASELINE_STORM_TIMES_KEY]
num_novel_examples = this_dict[NUM_NOVEL_EXAMPLES_KEY]
cnn_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(cnn_file_name)[0])
print('Reading CNN metadata from: "{0:s}"...'.format(cnn_metafile_name))
cnn_metadata_dict = cnn.read_model_metadata(cnn_metafile_name)
print(SEPARATOR_STRING)
baseline_predictor_matrices = testing_io.read_predictors_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_full_id_strings=baseline_full_id_strings,
desired_times_unix_sec=baseline_times_unix_sec,
option_dict=cnn_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY],
layer_operation_dicts=cnn_metadata_dict[cnn.LAYER_OPERATIONS_KEY])[
testing_io.INPUT_MATRICES_KEY]
print(SEPARATOR_STRING)
trial_predictor_matrices = testing_io.read_predictors_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_full_id_strings=trial_full_id_strings,
desired_times_unix_sec=trial_times_unix_sec,
option_dict=cnn_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY],
layer_operation_dicts=cnn_metadata_dict[cnn.LAYER_OPERATIONS_KEY])[
testing_io.INPUT_MATRICES_KEY]
print(SEPARATOR_STRING)
novelty_dict = novelty_detection.do_novelty_detection(
baseline_predictor_matrices=baseline_predictor_matrices,
trial_predictor_matrices=trial_predictor_matrices,
cnn_model_object=cnn_model_object,
cnn_feature_layer_name=cnn_feature_layer_name,
upconvnet_model_object=upconvnet_model_object,
num_novel_examples=num_novel_examples,
multipass=False,
percent_variance_to_keep=percent_variance_to_keep)
print(SEPARATOR_STRING)
print('Denormalizing inputs and outputs of novelty detection...')
cnn_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY][
trainval_io.SOUNDING_FIELDS_KEY] = None
novelty_dict[novelty_detection.BASELINE_MATRIX_KEY] = (
model_interpretation.denormalize_data(
list_of_input_matrices=baseline_predictor_matrices[[0]],
model_metadata_dict=cnn_metadata_dict))
novelty_dict[novelty_detection.TRIAL_MATRIX_KEY] = (
model_interpretation.denormalize_data(
list_of_input_matrices=trial_predictor_matrices[[0]],
model_metadata_dict=cnn_metadata_dict))
novelty_dict[novelty_detection.UPCONV_MATRIX_KEY] = (
model_interpretation.denormalize_data(
list_of_input_matrices=[
novelty_dict[novelty_detection.UPCONV_NORM_MATRIX_KEY]
],
model_metadata_dict=cnn_metadata_dict))[0]
novelty_dict.pop(novelty_detection.UPCONV_NORM_MATRIX_KEY)
novelty_dict[novelty_detection.UPCONV_SVD_MATRIX_KEY] = (
model_interpretation.denormalize_data(
list_of_input_matrices=[
novelty_dict[novelty_detection.UPCONV_NORM_SVD_MATRIX_KEY]
],
model_metadata_dict=cnn_metadata_dict))[0]
novelty_dict.pop(novelty_detection.UPCONV_NORM_SVD_MATRIX_KEY)
novelty_dict = novelty_detection.add_metadata(
novelty_dict=novelty_dict,
baseline_full_id_strings=baseline_full_id_strings,
baseline_times_unix_sec=baseline_times_unix_sec,
trial_full_id_strings=trial_full_id_strings,
trial_times_unix_sec=trial_times_unix_sec,
cnn_file_name=cnn_file_name,
upconvnet_file_name=upconvnet_file_name)
print('Writing results to: "{0:s}"...'.format(output_file_name))
novelty_detection.write_standard_file(novelty_dict=novelty_dict,
pickle_file_name=output_file_name)
def _run(model_file_name, init_function_name, storm_metafile_name,
num_examples, top_example_dir_name, component_type_string,
target_class, layer_name, neuron_indices, channel_index,
num_iterations, ideal_activation, learning_rate, l2_weight,
radar_constraint_weight, minmax_constraint_weight, output_file_name):
"""Runs backwards optimization on a trained CNN.
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param init_function_name: Same.
:param storm_metafile_name: Same.
:param num_examples: Same.
:param top_example_dir_name: Same.
:param component_type_string: Same.
:param target_class: Same.
:param layer_name: Same.
:param neuron_indices: Same.
:param channel_index: Same.
:param num_iterations: Same.
:param ideal_activation: Same.
:param learning_rate: Same.
:param l2_weight: Same.
:param radar_constraint_weight: Same.
:param minmax_constraint_weight: Same.
:param output_file_name: Same.
"""
if l2_weight <= 0:
l2_weight = None
if radar_constraint_weight <= 0:
radar_constraint_weight = None
if minmax_constraint_weight <= 0:
minmax_constraint_weight = None
if ideal_activation <= 0:
ideal_activation = None
if init_function_name in ['', 'None']:
init_function_name = None
model_interpretation.check_component_type(component_type_string)
model_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(model_file_name)[0])
print(
'Reading model metadata from: "{0:s}"...'.format(model_metafile_name))
model_metadata_dict = cnn.read_model_metadata(model_metafile_name)
input_matrices = None
init_function = None
full_storm_id_strings = None
storm_times_unix_sec = None
sounding_pressure_matrix_pa = None
if init_function_name is None:
print('Reading storm metadata from: "{0:s}"...'.format(
storm_metafile_name))
full_storm_id_strings, storm_times_unix_sec = (
tracking_io.read_ids_and_times(storm_metafile_name))
if 0 < num_examples < len(full_storm_id_strings):
full_storm_id_strings = full_storm_id_strings[:num_examples]
storm_times_unix_sec = storm_times_unix_sec[:num_examples]
example_dict = testing_io.read_predictors_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_full_id_strings=full_storm_id_strings,
desired_times_unix_sec=storm_times_unix_sec,
option_dict=model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY],
layer_operation_dicts=model_metadata_dict[
cnn.LAYER_OPERATIONS_KEY])
print(SEPARATOR_STRING)
input_matrices = example_dict[testing_io.INPUT_MATRICES_KEY]
sounding_pressure_matrix_pa = example_dict[
testing_io.SOUNDING_PRESSURES_KEY]
num_examples = input_matrices[0].shape[0]
else:
num_examples = 1
init_function = _create_initializer(
init_function_name=init_function_name,
model_metadata_dict=model_metadata_dict)
print('Reading model from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
output_matrices = None
initial_activations = numpy.full(num_examples, numpy.nan)
final_activations = numpy.full(num_examples, numpy.nan)
for i in range(num_examples):
if init_function_name is None:
this_init_arg = [a[[i], ...] for a in input_matrices]
else:
this_init_arg = init_function
if component_type_string == CLASS_COMPONENT_TYPE_STRING:
print((
'\nOptimizing {0:d}th of {1:d} images for target class {2:d}...'
).format(i + 1, num_examples, target_class))
this_result_dict = backwards_opt.optimize_input_for_class(
model_object=model_object,
target_class=target_class,
init_function_or_matrices=this_init_arg,
num_iterations=num_iterations,
learning_rate=learning_rate,
l2_weight=l2_weight,
radar_constraint_weight=radar_constraint_weight,
minmax_constraint_weight=minmax_constraint_weight,
model_metadata_dict=model_metadata_dict)
elif component_type_string == NEURON_COMPONENT_TYPE_STRING:
print((
'\nOptimizing {0:d}th of {1:d} images for neuron {2:s} in layer'
' "{3:s}"...').format(i + 1, num_examples, str(neuron_indices),
layer_name))
this_result_dict = backwards_opt.optimize_input_for_neuron(
model_object=model_object,
layer_name=layer_name,
neuron_indices=neuron_indices,
init_function_or_matrices=this_init_arg,
num_iterations=num_iterations,
learning_rate=learning_rate,
l2_weight=l2_weight,
ideal_activation=ideal_activation,
radar_constraint_weight=radar_constraint_weight,
minmax_constraint_weight=minmax_constraint_weight,
model_metadata_dict=model_metadata_dict)
else:
print(('\nOptimizing {0:d}th of {1:d} images for channel {2:d} in '
'layer "{3:s}"...').format(i + 1, num_examples,
channel_index, layer_name))
this_result_dict = backwards_opt.optimize_input_for_channel(
model_object=model_object,
layer_name=layer_name,
channel_index=channel_index,
init_function_or_matrices=this_init_arg,
stat_function_for_neuron_activations=K.max,
num_iterations=num_iterations,
learning_rate=learning_rate,
l2_weight=l2_weight,
ideal_activation=ideal_activation,
radar_constraint_weight=radar_constraint_weight,
minmax_constraint_weight=minmax_constraint_weight,
model_metadata_dict=model_metadata_dict)
initial_activations[i] = this_result_dict[
backwards_opt.INITIAL_ACTIVATION_KEY]
final_activations[i] = this_result_dict[
backwards_opt.FINAL_ACTIVATION_KEY]
these_output_matrices = this_result_dict[
backwards_opt.NORM_OUTPUT_MATRICES_KEY]
if output_matrices is None:
output_matrices = [None] * len(these_output_matrices)
for k in range(len(output_matrices)):
if output_matrices[k] is None:
output_matrices[k] = these_output_matrices[k] + 0.
else:
output_matrices[k] = numpy.concatenate(
(output_matrices[k], these_output_matrices[k]), axis=0)
if init_function_name is None:
continue
these_input_matrices = this_result_dict[
backwards_opt.NORM_INPUT_MATRICES_KEY]
if input_matrices is None:
input_matrices = [None] * len(these_input_matrices)
for k in range(len(input_matrices)):
if input_matrices[k] is None:
input_matrices[k] = these_input_matrices[k] + 0.
else:
input_matrices[k] = numpy.concatenate(
(input_matrices[k], these_input_matrices[k]), axis=0)
print(SEPARATOR_STRING)
training_option_dict = model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
print('Denormalizing input examples...')
input_matrices = trainval_io.separate_shear_and_reflectivity(
list_of_input_matrices=input_matrices,
training_option_dict=training_option_dict)
input_matrices = model_interpretation.denormalize_data(
list_of_input_matrices=input_matrices,
model_metadata_dict=model_metadata_dict)
print('Denormalizing optimized examples...')
output_matrices = trainval_io.separate_shear_and_reflectivity(
list_of_input_matrices=output_matrices,
training_option_dict=training_option_dict)
output_matrices = model_interpretation.denormalize_data(
list_of_input_matrices=output_matrices,
model_metadata_dict=model_metadata_dict)
print('Writing results to: "{0:s}"...'.format(output_file_name))
bwo_metadata_dict = backwards_opt.check_metadata(
component_type_string=component_type_string,
num_iterations=num_iterations,
learning_rate=learning_rate,
target_class=target_class,
layer_name=layer_name,
ideal_activation=ideal_activation,
neuron_indices=neuron_indices,
channel_index=channel_index,
l2_weight=l2_weight,
radar_constraint_weight=radar_constraint_weight,
minmax_constraint_weight=minmax_constraint_weight)
backwards_opt.write_standard_file(
pickle_file_name=output_file_name,
denorm_input_matrices=input_matrices,
denorm_output_matrices=output_matrices,
initial_activations=initial_activations,
final_activations=final_activations,
model_file_name=model_file_name,
metadata_dict=bwo_metadata_dict,
full_storm_id_strings=full_storm_id_strings,
storm_times_unix_sec=storm_times_unix_sec,
sounding_pressure_matrix_pa=sounding_pressure_matrix_pa)
def _run(cnn_file_name, upconvnet_file_name, top_example_dir_name,
baseline_storm_metafile_name, trial_storm_metafile_name,
num_baseline_examples, num_trial_examples, num_novel_examples,
cnn_feature_layer_name, percent_svd_variance_to_keep,
output_file_name):
"""Runs novelty detection.
This is effectively the main method.
:param cnn_file_name: See documentation at top of file.
:param upconvnet_file_name: Same.
:param top_example_dir_name: Same.
:param baseline_storm_metafile_name: Same.
:param trial_storm_metafile_name: Same.
:param num_baseline_examples: Same.
:param num_trial_examples: Same.
:param num_novel_examples: Same.
:param cnn_feature_layer_name: Same.
:param percent_svd_variance_to_keep: Same.
:param output_file_name: Same.
:raises: ValueError: if dimensions of first CNN input matrix != dimensions
of upconvnet output.
"""
print('Reading trained CNN from: "{0:s}"...'.format(cnn_file_name))
cnn_model_object = cnn.read_model(cnn_file_name)
cnn_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(cnn_file_name)[0]
)
print('Reading trained upconvnet from: "{0:s}"...'.format(
upconvnet_file_name))
upconvnet_model_object = cnn.read_model(upconvnet_file_name)
# ucn_output_dimensions = numpy.array(
# upconvnet_model_object.output.get_shape().as_list()[1:], dtype=int
# )
if isinstance(cnn_model_object.input, list):
first_cnn_input_tensor = cnn_model_object.input[0]
else:
first_cnn_input_tensor = cnn_model_object.input
cnn_input_dimensions = numpy.array(
first_cnn_input_tensor.get_shape().as_list()[1:], dtype=int
)
# if not numpy.array_equal(cnn_input_dimensions, ucn_output_dimensions):
# error_string = (
# 'Dimensions of first CNN input matrix ({0:s}) should equal '
# 'dimensions of upconvnet output ({1:s}).'
# ).format(str(cnn_input_dimensions), str(ucn_output_dimensions))
#
# raise ValueError(error_string)
print('Reading CNN metadata from: "{0:s}"...'.format(cnn_metafile_name))
cnn_metadata_dict = cnn.read_model_metadata(cnn_metafile_name)
print('Reading metadata for baseline examples from: "{0:s}"...'.format(
baseline_storm_metafile_name))
baseline_full_id_strings, baseline_times_unix_sec = (
tracking_io.read_ids_and_times(baseline_storm_metafile_name)
)
print('Reading metadata for trial examples from: "{0:s}"...'.format(
trial_storm_metafile_name))
trial_full_id_strings, trial_times_unix_sec = (
tracking_io.read_ids_and_times(trial_storm_metafile_name)
)
if 0 < num_baseline_examples < len(baseline_full_id_strings):
baseline_full_id_strings = baseline_full_id_strings[
:num_baseline_examples]
baseline_times_unix_sec = baseline_times_unix_sec[
:num_baseline_examples]
if 0 < num_trial_examples < len(trial_full_id_strings):
trial_full_id_strings = trial_full_id_strings[:num_trial_examples]
trial_times_unix_sec = trial_times_unix_sec[:num_trial_examples]
num_trial_examples = len(trial_full_id_strings)
if num_novel_examples <= 0:
num_novel_examples = num_trial_examples + 0
num_novel_examples = min([num_novel_examples, num_trial_examples])
print('Number of novel examples to find: {0:d}'.format(num_novel_examples))
bad_baseline_indices = tracking_utils.find_storm_objects(
all_id_strings=baseline_full_id_strings,
all_times_unix_sec=baseline_times_unix_sec,
id_strings_to_keep=trial_full_id_strings,
times_to_keep_unix_sec=trial_times_unix_sec, allow_missing=True)
print('Removing {0:d} trial examples from baseline set...'.format(
len(bad_baseline_indices)
))
baseline_times_unix_sec = numpy.delete(
baseline_times_unix_sec, bad_baseline_indices
)
baseline_full_id_strings = numpy.delete(
numpy.array(baseline_full_id_strings), bad_baseline_indices
)
baseline_full_id_strings = baseline_full_id_strings.tolist()
# num_baseline_examples = len(baseline_full_id_strings)
print(SEPARATOR_STRING)
list_of_baseline_input_matrices, _ = testing_io.read_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_full_id_strings=baseline_full_id_strings,
desired_times_unix_sec=baseline_times_unix_sec,
option_dict=cnn_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY],
list_of_layer_operation_dicts=cnn_metadata_dict[
cnn.LAYER_OPERATIONS_KEY]
)
print(SEPARATOR_STRING)
list_of_trial_input_matrices, _ = testing_io.read_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_full_id_strings=trial_full_id_strings,
desired_times_unix_sec=trial_times_unix_sec,
option_dict=cnn_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY],
list_of_layer_operation_dicts=cnn_metadata_dict[
cnn.LAYER_OPERATIONS_KEY]
)
print(SEPARATOR_STRING)
novelty_dict = novelty_detection.do_novelty_detection(
list_of_baseline_input_matrices=list_of_baseline_input_matrices,
list_of_trial_input_matrices=list_of_trial_input_matrices,
cnn_model_object=cnn_model_object,
cnn_feature_layer_name=cnn_feature_layer_name,
upconvnet_model_object=upconvnet_model_object,
num_novel_examples=num_novel_examples, multipass=False,
percent_svd_variance_to_keep=percent_svd_variance_to_keep)
print(SEPARATOR_STRING)
print('Adding metadata to novelty-detection results...')
novelty_dict = novelty_detection.add_metadata(
novelty_dict=novelty_dict,
baseline_full_id_strings=baseline_full_id_strings,
baseline_storm_times_unix_sec=baseline_times_unix_sec,
trial_full_id_strings=trial_full_id_strings,
trial_storm_times_unix_sec=trial_times_unix_sec,
cnn_file_name=cnn_file_name, upconvnet_file_name=upconvnet_file_name)
print('Denormalizing inputs and outputs of novelty detection...')
novelty_dict[novelty_detection.BASELINE_INPUTS_KEY] = (
model_interpretation.denormalize_data(
list_of_input_matrices=novelty_dict[
novelty_detection.BASELINE_INPUTS_KEY
],
model_metadata_dict=cnn_metadata_dict)
)
novelty_dict[novelty_detection.TRIAL_INPUTS_KEY] = (
model_interpretation.denormalize_data(
list_of_input_matrices=novelty_dict[
novelty_detection.TRIAL_INPUTS_KEY
],
model_metadata_dict=cnn_metadata_dict)
)
cnn_metadata_dict[
cnn.TRAINING_OPTION_DICT_KEY][trainval_io.SOUNDING_FIELDS_KEY] = None
novelty_dict[novelty_detection.NOVEL_IMAGES_UPCONV_KEY] = (
model_interpretation.denormalize_data(
list_of_input_matrices=[
novelty_dict[novelty_detection.NOVEL_IMAGES_UPCONV_KEY]
],
model_metadata_dict=cnn_metadata_dict)
)[0]
novelty_dict[novelty_detection.NOVEL_IMAGES_UPCONV_SVD_KEY] = (
model_interpretation.denormalize_data(
list_of_input_matrices=[
novelty_dict[novelty_detection.NOVEL_IMAGES_UPCONV_SVD_KEY]
],
model_metadata_dict=cnn_metadata_dict)
)[0]
print('Writing results to: "{0:s}"...'.format(output_file_name))
novelty_detection.write_standard_file(novelty_dict=novelty_dict,
pickle_file_name=output_file_name)
def _run(input_model_file_name, radar_field_name_by_channel,
layer_op_name_by_channel, min_height_by_channel_m_agl,
max_height_by_channel_m_agl, sounding_field_names,
normalization_type_string, normalization_param_file_name,
min_normalized_value, max_normalized_value, target_name,
shuffle_target, downsampling_classes, downsampling_fractions,
monitor_string, weight_loss_function, x_translations_pixels,
y_translations_pixels, ccw_rotation_angles_deg,
noise_standard_deviation, num_noisings, flip_in_x, flip_in_y,
top_training_dir_name, first_training_time_string,
last_training_time_string, num_examples_per_train_batch,
top_validation_dir_name, first_validation_time_string,
last_validation_time_string, num_examples_per_validn_batch,
num_epochs, num_training_batches_per_epoch,
num_validation_batches_per_epoch, output_dir_name):
"""Trains CNN with 2-D GridRad images.
This is effectively the main method.
:param input_model_file_name: See documentation at top of file.
:param radar_field_name_by_channel: Same.
:param layer_op_name_by_channel: Same.
:param min_height_by_channel_m_agl: Same.
:param max_height_by_channel_m_agl: Same.
:param sounding_field_names: Same.
:param normalization_type_string: Same.
:param normalization_param_file_name: Same.
:param min_normalized_value: Same.
:param max_normalized_value: Same.
:param target_name: Same.
:param shuffle_target: Same.
:param downsampling_classes: Same.
:param downsampling_fractions: Same.
:param monitor_string: Same.
:param weight_loss_function: Same.
:param x_translations_pixels: Same.
:param y_translations_pixels: Same.
:param ccw_rotation_angles_deg: Same.
:param noise_standard_deviation: Same.
:param num_noisings: Same.
:param flip_in_x: Same.
:param flip_in_y: Same.
:param top_training_dir_name: Same.
:param first_training_time_string: Same.
:param last_training_time_string: Same.
:param num_examples_per_train_batch: Same.
:param top_validation_dir_name: Same.
:param first_validation_time_string: Same.
:param last_validation_time_string: Same.
:param num_examples_per_validn_batch: Same.
:param num_epochs: Same.
:param num_training_batches_per_epoch: Same.
:param num_validation_batches_per_epoch: Same.
:param output_dir_name: Same.
"""
# Process input args.
first_training_time_unix_sec = time_conversion.string_to_unix_sec(
first_training_time_string, TIME_FORMAT)
last_training_time_unix_sec = time_conversion.string_to_unix_sec(
last_training_time_string, TIME_FORMAT)
if top_validation_dir_name in ['', 'None']:
top_validation_dir_name = None
num_validation_batches_per_epoch = 0
first_validation_time_unix_sec = 0
last_validation_time_unix_sec = 0
else:
first_validation_time_unix_sec = time_conversion.string_to_unix_sec(
first_validation_time_string, TIME_FORMAT)
last_validation_time_unix_sec = time_conversion.string_to_unix_sec(
last_validation_time_string, TIME_FORMAT)
if sounding_field_names[0] in ['', 'None']:
sounding_field_names = None
if len(downsampling_classes) > 1:
class_to_sampling_fraction_dict = dict(
list(zip(downsampling_classes, downsampling_fractions)))
else:
class_to_sampling_fraction_dict = None
if (len(x_translations_pixels) == 1
and x_translations_pixels + y_translations_pixels == 0):
x_translations_pixels = None
y_translations_pixels = None
if len(ccw_rotation_angles_deg) == 1 and ccw_rotation_angles_deg[0] == 0:
ccw_rotation_angles_deg = None
if num_noisings <= 0:
num_noisings = 0
noise_standard_deviation = None
num_channels = len(radar_field_name_by_channel)
expected_dimensions = numpy.array([num_channels], dtype=int)
error_checking.assert_is_numpy_array(numpy.array(layer_op_name_by_channel),
exact_dimensions=expected_dimensions)
error_checking.assert_is_numpy_array(min_height_by_channel_m_agl,
exact_dimensions=expected_dimensions)
error_checking.assert_is_numpy_array(max_height_by_channel_m_agl,
exact_dimensions=expected_dimensions)
list_of_layer_operation_dicts = [{}] * num_channels
for m in range(num_channels):
list_of_layer_operation_dicts[m] = {
input_examples.RADAR_FIELD_KEY: radar_field_name_by_channel[m],
input_examples.OPERATION_NAME_KEY: layer_op_name_by_channel[m],
input_examples.MIN_HEIGHT_KEY: min_height_by_channel_m_agl[m],
input_examples.MAX_HEIGHT_KEY: max_height_by_channel_m_agl[m]
}
# Set output locations.
file_system_utils.mkdir_recursive_if_necessary(
directory_name=output_dir_name)
output_model_file_name = '{0:s}/model.h5'.format(output_dir_name)
history_file_name = '{0:s}/model_history.csv'.format(output_dir_name)
tensorboard_dir_name = '{0:s}/tensorboard'.format(output_dir_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(output_dir_name)
# Find training and validation files.
training_file_names = input_examples.find_many_example_files(
top_directory_name=top_training_dir_name,
shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER,
raise_error_if_any_missing=False)
if top_validation_dir_name is None:
validation_file_names = []
else:
validation_file_names = input_examples.find_many_example_files(
top_directory_name=top_validation_dir_name,
shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER,
raise_error_if_any_missing=False)
# Read architecture.
print(
'Reading architecture from: "{0:s}"...'.format(input_model_file_name))
model_object = cnn.read_model(input_model_file_name)
# model_object = clone_model(model_object)
# TODO(thunderhoser): This is a HACK.
model_object.compile(loss=keras.losses.binary_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=cnn_setup.DEFAULT_METRIC_FUNCTION_LIST)
print(SEPARATOR_STRING)
model_object.summary()
print(SEPARATOR_STRING)
print(K.eval(model_object.get_layer(name='radar_conv2d_2').weights[0]))
print(SEPARATOR_STRING)
# Write metadata.
metadata_dict = {
cnn.NUM_EPOCHS_KEY: num_epochs,
cnn.NUM_TRAINING_BATCHES_KEY: num_training_batches_per_epoch,
cnn.NUM_VALIDATION_BATCHES_KEY: num_validation_batches_per_epoch,
cnn.MONITOR_STRING_KEY: monitor_string,
cnn.WEIGHT_LOSS_FUNCTION_KEY: weight_loss_function,
cnn.CONV_2D3D_KEY: False,
cnn.VALIDATION_FILES_KEY: validation_file_names,
cnn.FIRST_VALIDN_TIME_KEY: first_validation_time_unix_sec,
cnn.LAST_VALIDN_TIME_KEY: last_validation_time_unix_sec,
cnn.NUM_EX_PER_VALIDN_BATCH_KEY: num_examples_per_validn_batch
}
input_tensor = model_object.input
if isinstance(input_tensor, list):
input_tensor = input_tensor[0]
num_grid_rows = input_tensor.get_shape().as_list()[1]
num_grid_columns = input_tensor.get_shape().as_list()[2]
training_option_dict = {
trainval_io.EXAMPLE_FILES_KEY: training_file_names,
trainval_io.TARGET_NAME_KEY: target_name,
trainval_io.SHUFFLE_TARGET_KEY: shuffle_target,
trainval_io.FIRST_STORM_TIME_KEY: first_training_time_unix_sec,
trainval_io.LAST_STORM_TIME_KEY: last_training_time_unix_sec,
trainval_io.NUM_EXAMPLES_PER_BATCH_KEY: num_examples_per_train_batch,
trainval_io.SOUNDING_FIELDS_KEY: sounding_field_names,
trainval_io.SOUNDING_HEIGHTS_KEY: SOUNDING_HEIGHTS_M_AGL,
trainval_io.NUM_ROWS_KEY: num_grid_rows,
trainval_io.NUM_COLUMNS_KEY: num_grid_columns,
trainval_io.NORMALIZATION_TYPE_KEY: normalization_type_string,
trainval_io.NORMALIZATION_FILE_KEY: normalization_param_file_name,
trainval_io.MIN_NORMALIZED_VALUE_KEY: min_normalized_value,
trainval_io.MAX_NORMALIZED_VALUE_KEY: max_normalized_value,
trainval_io.BINARIZE_TARGET_KEY: False,
trainval_io.SAMPLING_FRACTIONS_KEY: class_to_sampling_fraction_dict,
trainval_io.LOOP_ONCE_KEY: False,
trainval_io.X_TRANSLATIONS_KEY: x_translations_pixels,
trainval_io.Y_TRANSLATIONS_KEY: y_translations_pixels,
trainval_io.ROTATION_ANGLES_KEY: ccw_rotation_angles_deg,
trainval_io.NOISE_STDEV_KEY: noise_standard_deviation,
trainval_io.NUM_NOISINGS_KEY: num_noisings,
trainval_io.FLIP_X_KEY: flip_in_x,
trainval_io.FLIP_Y_KEY: flip_in_y
}
print('Writing metadata to: "{0:s}"...'.format(model_metafile_name))
cnn.write_model_metadata(
pickle_file_name=model_metafile_name,
metadata_dict=metadata_dict,
training_option_dict=training_option_dict,
list_of_layer_operation_dicts=list_of_layer_operation_dicts)
cnn.train_cnn_gridrad_2d_reduced(
model_object=model_object,
model_file_name=output_model_file_name,
history_file_name=history_file_name,
tensorboard_dir_name=tensorboard_dir_name,
num_epochs=num_epochs,
num_training_batches_per_epoch=num_training_batches_per_epoch,
training_option_dict=training_option_dict,
list_of_layer_operation_dicts=list_of_layer_operation_dicts,
monitor_string=monitor_string,
weight_loss_function=weight_loss_function,
num_validation_batches_per_epoch=num_validation_batches_per_epoch,
validation_file_names=validation_file_names,
first_validn_time_unix_sec=first_validation_time_unix_sec,
last_validn_time_unix_sec=last_validation_time_unix_sec,
num_examples_per_validn_batch=num_examples_per_validn_batch)
def _run(input_model_file_name, radar_field_names, sounding_field_names,
normalization_type_string, normalization_param_file_name,
min_normalized_value, max_normalized_value, downsampling_keys,
downsampling_fractions, monitor_string, weight_loss_function,
refl_masking_threshold_dbz, x_translations_pixels,
y_translations_pixels, ccw_rotation_angles_deg,
noise_standard_deviation, num_noisings, flip_in_x, flip_in_y,
top_training_dir_name, first_training_time_string,
last_training_time_string, top_validation_dir_name,
first_validation_time_string, last_validation_time_string,
num_examples_per_batch, num_epochs, num_training_batches_per_epoch,
num_validation_batches_per_epoch, output_dir_name):
"""Trains CNN with native (3-D) GridRad images.
This is effectively the main method.
:param input_model_file_name: See documentation at top of file.
:param radar_field_names: Same.
:param sounding_field_names: Same.
:param normalization_type_string: Same.
:param normalization_param_file_name: Same.
:param min_normalized_value: Same.
:param max_normalized_value: Same.
:param downsampling_keys: Same.
:param downsampling_fractions: Same.
:param monitor_string: Same.
:param weight_loss_function: Same.
:param refl_masking_threshold_dbz: Same.
:param x_translations_pixels: Same.
:param y_translations_pixels: Same.
:param ccw_rotation_angles_deg: Same.
:param noise_standard_deviation: Same.
:param num_noisings: Same.
:param flip_in_x: Same.
:param flip_in_y: Same.
:param top_training_dir_name: Same.
:param first_training_time_string: Same.
:param last_training_time_string: Same.
:param top_validation_dir_name: Same.
:param first_validation_time_string: Same.
:param last_validation_time_string: Same.
:param num_examples_per_batch: Same.
:param num_epochs: Same.
:param num_training_batches_per_epoch: Same.
:param num_validation_batches_per_epoch: Same.
:param output_dir_name: Same.
"""
# Process input args.
first_training_time_unix_sec = time_conversion.string_to_unix_sec(
first_training_time_string, TIME_FORMAT)
last_training_time_unix_sec = time_conversion.string_to_unix_sec(
last_training_time_string, TIME_FORMAT)
first_validation_time_unix_sec = time_conversion.string_to_unix_sec(
first_validation_time_string, TIME_FORMAT)
last_validation_time_unix_sec = time_conversion.string_to_unix_sec(
last_validation_time_string, TIME_FORMAT)
if sounding_field_names[0] in ['', 'None']:
sounding_field_names = None
if len(downsampling_keys) > 1:
class_to_sampling_fraction_dict = dict(
zip(downsampling_keys, downsampling_fractions))
else:
class_to_sampling_fraction_dict = None
if (len(x_translations_pixels) == 1
and x_translations_pixels + y_translations_pixels == 0):
x_translations_pixels = None
y_translations_pixels = None
if len(ccw_rotation_angles_deg) == 1 and ccw_rotation_angles_deg[0] == 0:
ccw_rotation_angles_deg = None
if num_noisings <= 0:
num_noisings = 0
noise_standard_deviation = None
# Set output locations.
file_system_utils.mkdir_recursive_if_necessary(
directory_name=output_dir_name)
output_model_file_name = '{0:s}/model.h5'.format(output_dir_name)
history_file_name = '{0:s}/model_history.csv'.format(output_dir_name)
tensorboard_dir_name = '{0:s}/tensorboard'.format(output_dir_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(output_dir_name)
# Find training and validation files.
training_file_names = input_examples.find_many_example_files(
top_directory_name=top_training_dir_name,
shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER,
raise_error_if_any_missing=False)
validation_file_names = input_examples.find_many_example_files(
top_directory_name=top_validation_dir_name,
shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER,
raise_error_if_any_missing=False)
# Read architecture.
print 'Reading architecture from: "{0:s}"...'.format(input_model_file_name)
model_object = cnn.read_model(input_model_file_name)
model_object = keras.models.clone_model(model_object)
# TODO(thunderhoser): This is a HACK.
model_object.compile(loss=keras.losses.binary_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=cnn_setup.DEFAULT_METRIC_FUNCTION_LIST)
print SEPARATOR_STRING
model_object.summary()
print SEPARATOR_STRING
# Write metadata.
this_example_dict = input_examples.read_example_file(
netcdf_file_name=training_file_names[0], metadata_only=True)
target_name = this_example_dict[input_examples.TARGET_NAME_KEY]
metadata_dict = {
cnn.TARGET_NAME_KEY: target_name,
cnn.NUM_EPOCHS_KEY: num_epochs,
cnn.NUM_TRAINING_BATCHES_KEY: num_training_batches_per_epoch,
cnn.NUM_VALIDATION_BATCHES_KEY: num_validation_batches_per_epoch,
cnn.MONITOR_STRING_KEY: monitor_string,
cnn.WEIGHT_LOSS_FUNCTION_KEY: weight_loss_function,
cnn.USE_2D3D_CONVOLUTION_KEY: False,
cnn.VALIDATION_FILES_KEY: validation_file_names,
cnn.FIRST_VALIDN_TIME_KEY: first_validation_time_unix_sec,
cnn.LAST_VALIDN_TIME_KEY: last_validation_time_unix_sec
}
input_tensor = model_object.input
if isinstance(input_tensor, list):
input_tensor = input_tensor[0]
num_grid_rows = input_tensor.get_shape().as_list()[1]
num_grid_columns = input_tensor.get_shape().as_list()[2]
training_option_dict = {
trainval_io.EXAMPLE_FILES_KEY: training_file_names,
trainval_io.FIRST_STORM_TIME_KEY: first_training_time_unix_sec,
trainval_io.LAST_STORM_TIME_KEY: last_training_time_unix_sec,
trainval_io.NUM_EXAMPLES_PER_BATCH_KEY: num_examples_per_batch,
trainval_io.RADAR_FIELDS_KEY: radar_field_names,
trainval_io.RADAR_HEIGHTS_KEY: RADAR_HEIGHTS_M_AGL,
trainval_io.SOUNDING_FIELDS_KEY: sounding_field_names,
trainval_io.SOUNDING_HEIGHTS_KEY: SOUNDING_HEIGHTS_M_AGL,
trainval_io.NUM_ROWS_KEY: num_grid_rows,
trainval_io.NUM_COLUMNS_KEY: num_grid_columns,
trainval_io.NORMALIZATION_TYPE_KEY: normalization_type_string,
trainval_io.NORMALIZATION_FILE_KEY: normalization_param_file_name,
trainval_io.MIN_NORMALIZED_VALUE_KEY: min_normalized_value,
trainval_io.MAX_NORMALIZED_VALUE_KEY: max_normalized_value,
trainval_io.BINARIZE_TARGET_KEY: False,
trainval_io.SAMPLING_FRACTIONS_KEY: class_to_sampling_fraction_dict,
trainval_io.LOOP_ONCE_KEY: False,
trainval_io.REFLECTIVITY_MASK_KEY: refl_masking_threshold_dbz,
trainval_io.X_TRANSLATIONS_KEY: x_translations_pixels,
trainval_io.Y_TRANSLATIONS_KEY: y_translations_pixels,
trainval_io.ROTATION_ANGLES_KEY: ccw_rotation_angles_deg,
trainval_io.NOISE_STDEV_KEY: noise_standard_deviation,
trainval_io.NUM_NOISINGS_KEY: num_noisings,
trainval_io.FLIP_X_KEY: flip_in_x,
trainval_io.FLIP_Y_KEY: flip_in_y
}
print 'Writing metadata to: "{0:s}"...'.format(model_metafile_name)
cnn.write_model_metadata(pickle_file_name=model_metafile_name,
metadata_dict=metadata_dict,
training_option_dict=training_option_dict)
cnn.train_cnn_2d_or_3d(
model_object=model_object,
model_file_name=output_model_file_name,
history_file_name=history_file_name,
tensorboard_dir_name=tensorboard_dir_name,
num_epochs=num_epochs,
num_training_batches_per_epoch=num_training_batches_per_epoch,
training_option_dict=training_option_dict,
monitor_string=monitor_string,
weight_loss_function=weight_loss_function,
num_validation_batches_per_epoch=num_validation_batches_per_epoch,
validation_file_names=validation_file_names,
first_validn_time_unix_sec=first_validation_time_unix_sec,
last_validn_time_unix_sec=last_validation_time_unix_sec)
def _run(model_file_name, top_example_dir_name, storm_metafile_name,
output_dir_name):
"""Uses trained CNN to make predictions for specific examples.
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param top_example_dir_name: Same.
:param storm_metafile_name: Same.
:param output_dir_name: Same.
:raises: ValueError: if the model does multi-class classification.
"""
print('Reading CNN from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
num_output_neurons = (
model_object.layers[-1].output.get_shape().as_list()[-1]
)
if num_output_neurons > 2:
error_string = (
'The model has {0:d} output neurons, which suggests {0:d}-class '
'classification. This script handles only binary classification.'
).format(num_output_neurons)
raise ValueError(error_string)
soundings_only = False
if isinstance(model_object.input, list):
list_of_input_tensors = model_object.input
else:
list_of_input_tensors = [model_object.input]
if len(list_of_input_tensors) == 1:
these_spatial_dim = numpy.array(
list_of_input_tensors[0].get_shape().as_list()[1:-1], dtype=int
)
soundings_only = len(these_spatial_dim) == 1
cnn_metafile_name = cnn.find_metafile(
model_file_name=model_file_name, raise_error_if_missing=True
)
print('Reading CNN metadata from: "{0:s}"...'.format(cnn_metafile_name))
cnn_metadata_dict = cnn.read_model_metadata(cnn_metafile_name)
print('Reading storm metadata from: "{0:s}"...'.format(storm_metafile_name))
desired_full_id_strings, desired_times_unix_sec = (
tracking_io.read_ids_and_times(storm_metafile_name)
)
unique_spc_date_strings = list(set([
time_conversion.time_to_spc_date_string(t)
for t in desired_times_unix_sec
]))
example_file_names = [
input_examples.find_example_file(
top_directory_name=top_example_dir_name, shuffled=False,
spc_date_string=d, raise_error_if_missing=True
) for d in unique_spc_date_strings
]
first_spc_date_string = time_conversion.time_to_spc_date_string(
numpy.min(desired_times_unix_sec)
)
last_spc_date_string = time_conversion.time_to_spc_date_string(
numpy.max(desired_times_unix_sec)
)
training_option_dict = cnn_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
training_option_dict[trainval_io.EXAMPLE_FILES_KEY] = example_file_names
training_option_dict[trainval_io.FIRST_STORM_TIME_KEY] = (
time_conversion.get_start_of_spc_date(first_spc_date_string)
)
training_option_dict[trainval_io.LAST_STORM_TIME_KEY] = (
time_conversion.get_end_of_spc_date(last_spc_date_string)
)
training_option_dict[trainval_io.NUM_EXAMPLES_PER_BATCH_KEY] = (
NUM_EXAMPLES_PER_BATCH
)
if soundings_only:
generator_object = testing_io.sounding_generator(
option_dict=training_option_dict,
desired_full_id_strings=desired_full_id_strings,
desired_times_unix_sec=desired_times_unix_sec)
elif cnn_metadata_dict[cnn.LAYER_OPERATIONS_KEY] is not None:
generator_object = testing_io.gridrad_generator_2d_reduced(
option_dict=training_option_dict,
desired_full_id_strings=desired_full_id_strings,
desired_times_unix_sec=desired_times_unix_sec,
list_of_operation_dicts=cnn_metadata_dict[
cnn.LAYER_OPERATIONS_KEY]
)
elif cnn_metadata_dict[cnn.CONV_2D3D_KEY]:
generator_object = testing_io.myrorss_generator_2d3d(
option_dict=training_option_dict,
desired_full_id_strings=desired_full_id_strings,
desired_times_unix_sec=desired_times_unix_sec)
else:
generator_object = testing_io.generator_2d_or_3d(
option_dict=training_option_dict,
desired_full_id_strings=desired_full_id_strings,
desired_times_unix_sec=desired_times_unix_sec)
include_soundings = (
training_option_dict[trainval_io.SOUNDING_FIELDS_KEY] is not None
)
full_storm_id_strings = []
storm_times_unix_sec = numpy.array([], dtype=int)
observed_labels = numpy.array([], dtype=int)
class_probability_matrix = None
while True:
try:
this_storm_object_dict = next(generator_object)
print(SEPARATOR_STRING)
except StopIteration:
break
full_storm_id_strings += this_storm_object_dict[testing_io.FULL_IDS_KEY]
storm_times_unix_sec = numpy.concatenate((
storm_times_unix_sec,
this_storm_object_dict[testing_io.STORM_TIMES_KEY]
))
observed_labels = numpy.concatenate((
observed_labels, this_storm_object_dict[testing_io.TARGET_ARRAY_KEY]
))
if soundings_only:
these_predictor_matrices = [
this_storm_object_dict[testing_io.SOUNDING_MATRIX_KEY]
]
else:
these_predictor_matrices = this_storm_object_dict[
testing_io.INPUT_MATRICES_KEY]
if include_soundings:
this_sounding_matrix = these_predictor_matrices[-1]
else:
this_sounding_matrix = None
if soundings_only:
this_probability_matrix = cnn.apply_cnn_soundings_only(
model_object=model_object, sounding_matrix=this_sounding_matrix,
verbose=True)
elif cnn_metadata_dict[cnn.CONV_2D3D_KEY]:
if training_option_dict[trainval_io.UPSAMPLE_REFLECTIVITY_KEY]:
this_probability_matrix = cnn.apply_2d_or_3d_cnn(
model_object=model_object,
radar_image_matrix=these_predictor_matrices[0],
sounding_matrix=this_sounding_matrix, verbose=True)
else:
this_probability_matrix = cnn.apply_2d3d_cnn(
model_object=model_object,
reflectivity_matrix_dbz=these_predictor_matrices[0],
azimuthal_shear_matrix_s01=these_predictor_matrices[1],
sounding_matrix=this_sounding_matrix, verbose=True)
else:
this_probability_matrix = cnn.apply_2d_or_3d_cnn(
model_object=model_object,
radar_image_matrix=these_predictor_matrices[0],
sounding_matrix=this_sounding_matrix, verbose=True)
print(SEPARATOR_STRING)
if class_probability_matrix is None:
class_probability_matrix = this_probability_matrix + 0.
else:
class_probability_matrix = numpy.concatenate(
(class_probability_matrix, this_probability_matrix), axis=0
)
output_file_name = prediction_io.find_ungridded_file(
directory_name=output_dir_name, raise_error_if_missing=False)
print('Writing results to: "{0:s}"...'.format(output_file_name))
prediction_io.write_ungridded_predictions(
netcdf_file_name=output_file_name,
class_probability_matrix=class_probability_matrix,
observed_labels=observed_labels, storm_ids=full_storm_id_strings,
storm_times_unix_sec=storm_times_unix_sec,
target_name=training_option_dict[trainval_io.TARGET_NAME_KEY],
model_file_name=model_file_name
)
def _run(upconvnet_file_name, storm_metafile_name, num_examples,
top_example_dir_name, top_output_dir_name):
"""Plots upconvnet reconstructions of many examples (storm objects).
This is effectively the main method.
:param upconvnet_file_name: See documentation at top of file.
:param storm_metafile_name: Same.
:param num_examples: Same.
:param top_example_dir_name: Same.
:param top_output_dir_name: Same.
"""
print 'Reading trained upconvnet from: "{0:s}"...'.format(
upconvnet_file_name)
upconvnet_model_object = cnn.read_model(upconvnet_file_name)
upconvnet_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(upconvnet_file_name)[0]
)
print 'Reading upconvnet metadata from: "{0:s}"...'.format(
upconvnet_metafile_name)
upconvnet_metadata_dict = upconvnet.read_model_metadata(
upconvnet_metafile_name)
cnn_file_name = upconvnet_metadata_dict[upconvnet.CNN_FILE_KEY]
print 'Reading trained CNN from: "{0:s}"...'.format(cnn_file_name)
cnn_model_object = cnn.read_model(cnn_file_name)
cnn_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(cnn_file_name)[0]
)
print 'Reading CNN metadata from: "{0:s}"...'.format(cnn_metafile_name)
cnn_metadata_dict = cnn.read_model_metadata(cnn_metafile_name)
training_option_dict = cnn_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
print 'Reading storm IDs and times from: "{0:s}"...'.format(
storm_metafile_name)
storm_ids, storm_times_unix_sec = tracking_io.read_ids_and_times(
storm_metafile_name)
if 0 < num_examples < len(storm_ids):
storm_ids = storm_ids[:num_examples]
storm_times_unix_sec = storm_times_unix_sec[:num_examples]
print SEPARATOR_STRING
list_of_predictor_matrices = testing_io.read_specific_examples(
desired_storm_ids=storm_ids,
desired_times_unix_sec=storm_times_unix_sec,
option_dict=training_option_dict,
top_example_dir_name=top_example_dir_name,
list_of_layer_operation_dicts=cnn_metadata_dict[
cnn.LAYER_OPERATIONS_KEY]
)[0]
print SEPARATOR_STRING
actual_radar_matrix = list_of_predictor_matrices[0]
have_soundings = training_option_dict[trainval_io.SOUNDING_FIELDS_KEY]
if have_soundings:
sounding_matrix = list_of_predictor_matrices[-1]
else:
sounding_matrix = None
feature_matrix = cnn.apply_2d_or_3d_cnn(
model_object=cnn_model_object, radar_image_matrix=actual_radar_matrix,
sounding_matrix=sounding_matrix, verbose=True, return_features=True,
feature_layer_name=upconvnet_metadata_dict[
upconvnet.CNN_FEATURE_LAYER_KEY]
)
print '\n'
reconstructed_radar_matrix = upconvnet.apply_upconvnet(
model_object=upconvnet_model_object, feature_matrix=feature_matrix,
verbose=True)
print '\n'
print 'Denormalizing actual and reconstructed radar images...'
cnn_metadata_dict[
cnn.TRAINING_OPTION_DICT_KEY][trainval_io.SOUNDING_FIELDS_KEY] = None
actual_radar_matrix = model_interpretation.denormalize_data(
list_of_input_matrices=[actual_radar_matrix],
model_metadata_dict=cnn_metadata_dict
)[0]
reconstructed_radar_matrix = model_interpretation.denormalize_data(
list_of_input_matrices=[reconstructed_radar_matrix],
model_metadata_dict=cnn_metadata_dict
)[0]
print SEPARATOR_STRING
actual_output_dir_name = '{0:s}/actual_images'.format(top_output_dir_name)
file_system_utils.mkdir_recursive_if_necessary(
directory_name=actual_output_dir_name)
# TODO(thunderhoser): Calling a method in another script is hacky. If this
# method is going to be reused, should be in a module.
plot_input_examples.plot_examples(
list_of_predictor_matrices=[actual_radar_matrix], storm_ids=storm_ids,
storm_times_unix_sec=storm_times_unix_sec,
model_metadata_dict=cnn_metadata_dict,
output_dir_name=actual_output_dir_name)
print SEPARATOR_STRING
reconstructed_output_dir_name = '{0:s}/reconstructed_images'.format(
top_output_dir_name)
file_system_utils.mkdir_recursive_if_necessary(
directory_name=reconstructed_output_dir_name)
plot_input_examples.plot_examples(
list_of_predictor_matrices=[reconstructed_radar_matrix],
storm_ids=storm_ids, storm_times_unix_sec=storm_times_unix_sec,
model_metadata_dict=cnn_metadata_dict,
output_dir_name=reconstructed_output_dir_name)
def _run(input_model_file_name, sounding_field_names, normalization_type_string,
normalization_param_file_name, min_normalized_value,
max_normalized_value, target_name, downsampling_classes,
downsampling_fractions, monitor_string, weight_loss_function,
top_training_dir_name, first_training_time_string,
last_training_time_string, num_examples_per_train_batch,
top_validation_dir_name, first_validation_time_string,
last_validation_time_string, num_examples_per_validn_batch, num_epochs,
num_training_batches_per_epoch, num_validation_batches_per_epoch,
output_dir_name):
"""Trains CNN with soundings only.
This is effectively the main method.
:param input_model_file_name: See documentation at top of file.
:param sounding_field_names: Same.
:param normalization_type_string: Same.
:param normalization_param_file_name: Same.
:param min_normalized_value: Same.
:param max_normalized_value: Same.
:param target_name: Same.
:param downsampling_classes: Same.
:param downsampling_fractions: Same.
:param monitor_string: Same.
:param weight_loss_function: Same.
:param top_training_dir_name: Same.
:param first_training_time_string: Same.
:param last_training_time_string: Same.
:param num_examples_per_train_batch: Same.
:param top_validation_dir_name: Same.
:param first_validation_time_string: Same.
:param last_validation_time_string: Same.
:param num_examples_per_validn_batch: Same.
:param num_epochs: Same.
:param num_training_batches_per_epoch: Same.
:param num_validation_batches_per_epoch: Same.
:param output_dir_name: Same.
"""
# Process input args.
first_training_time_unix_sec = time_conversion.string_to_unix_sec(
first_training_time_string, TIME_FORMAT)
last_training_time_unix_sec = time_conversion.string_to_unix_sec(
last_training_time_string, TIME_FORMAT)
first_validation_time_unix_sec = time_conversion.string_to_unix_sec(
first_validation_time_string, TIME_FORMAT)
last_validation_time_unix_sec = time_conversion.string_to_unix_sec(
last_validation_time_string, TIME_FORMAT)
if len(downsampling_classes) > 1:
downsampling_dict = dict(list(zip(
downsampling_classes, downsampling_fractions
)))
else:
downsampling_dict = None
# Set output locations.
file_system_utils.mkdir_recursive_if_necessary(
directory_name=output_dir_name)
output_model_file_name = '{0:s}/model.h5'.format(output_dir_name)
history_file_name = '{0:s}/model_history.csv'.format(output_dir_name)
tensorboard_dir_name = '{0:s}/tensorboard'.format(output_dir_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(output_dir_name)
# Find training and validation files.
training_file_names = input_examples.find_many_example_files(
top_directory_name=top_training_dir_name, shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER, raise_error_if_any_missing=False)
validation_file_names = input_examples.find_many_example_files(
top_directory_name=top_validation_dir_name, shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER, raise_error_if_any_missing=False)
# Read architecture.
print('Reading architecture from: "{0:s}"...'.format(input_model_file_name))
model_object = cnn.read_model(input_model_file_name)
# model_object = keras.models.clone_model(model_object)
# TODO(thunderhoser): This is a HACK.
model_object.compile(
loss=keras.losses.binary_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=cnn_setup.DEFAULT_METRIC_FUNCTION_LIST)
print(SEPARATOR_STRING)
model_object.summary()
print(SEPARATOR_STRING)
# Write metadata.
metadata_dict = {
cnn.NUM_EPOCHS_KEY: num_epochs,
cnn.NUM_TRAINING_BATCHES_KEY: num_training_batches_per_epoch,
cnn.NUM_VALIDATION_BATCHES_KEY: num_validation_batches_per_epoch,
cnn.MONITOR_STRING_KEY: monitor_string,
cnn.WEIGHT_LOSS_FUNCTION_KEY: weight_loss_function,
cnn.CONV_2D3D_KEY: False,
cnn.VALIDATION_FILES_KEY: validation_file_names,
cnn.FIRST_VALIDN_TIME_KEY: first_validation_time_unix_sec,
cnn.LAST_VALIDN_TIME_KEY: last_validation_time_unix_sec,
cnn.NUM_EX_PER_VALIDN_BATCH_KEY: num_examples_per_validn_batch
}
training_option_dict = {
trainval_io.EXAMPLE_FILES_KEY: training_file_names,
trainval_io.NUM_EXAMPLES_PER_BATCH_KEY: num_examples_per_train_batch,
trainval_io.FIRST_STORM_TIME_KEY: first_training_time_unix_sec,
trainval_io.LAST_STORM_TIME_KEY: last_training_time_unix_sec,
trainval_io.SOUNDING_FIELDS_KEY: sounding_field_names,
trainval_io.SOUNDING_HEIGHTS_KEY: SOUNDING_HEIGHTS_M_AGL,
trainval_io.NORMALIZATION_TYPE_KEY: normalization_type_string,
trainval_io.NORMALIZATION_FILE_KEY: normalization_param_file_name,
trainval_io.MIN_NORMALIZED_VALUE_KEY: min_normalized_value,
trainval_io.MAX_NORMALIZED_VALUE_KEY: max_normalized_value,
trainval_io.TARGET_NAME_KEY: target_name,
trainval_io.BINARIZE_TARGET_KEY: False,
trainval_io.SAMPLING_FRACTIONS_KEY: downsampling_dict,
trainval_io.LOOP_ONCE_KEY: False
}
print('Writing metadata to: "{0:s}"...'.format(model_metafile_name))
cnn.write_model_metadata(
pickle_file_name=model_metafile_name, metadata_dict=metadata_dict,
training_option_dict=training_option_dict)
cnn.train_cnn_with_soundings(
model_object=model_object, model_file_name=output_model_file_name,
history_file_name=history_file_name,
tensorboard_dir_name=tensorboard_dir_name, num_epochs=num_epochs,
num_training_batches_per_epoch=num_training_batches_per_epoch,
training_option_dict=training_option_dict,
monitor_string=monitor_string,
weight_loss_function=weight_loss_function,
num_validation_batches_per_epoch=num_validation_batches_per_epoch,
validation_file_names=validation_file_names,
first_validn_time_unix_sec=first_validation_time_unix_sec,
last_validn_time_unix_sec=last_validation_time_unix_sec,
num_examples_per_validn_batch=num_examples_per_validn_batch)
def _run(model_file_name, target_class, target_layer_name,
top_example_dir_name, storm_metafile_name, num_examples,
randomize_weights, cascading_random, output_file_name):
"""Runs Grad-CAM (gradient-weighted class-activation maps).
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param target_class: Same.
:param target_layer_name: Same.
:param top_example_dir_name: Same.
:param storm_metafile_name: Same.
:param num_examples: Same.
:param randomize_weights: Same.
:param cascading_random: Same.
:param output_file_name: Same.
"""
file_system_utils.mkdir_recursive_if_necessary(file_name=output_file_name)
# Read model and metadata.
print('Reading model from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(model_file_name)[0])
print(
'Reading model metadata from: "{0:s}"...'.format(model_metafile_name))
model_metadata_dict = cnn.read_model_metadata(model_metafile_name)
training_option_dict = model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
training_option_dict[trainval_io.REFLECTIVITY_MASK_KEY] = None
output_dir_name, pathless_output_file_name = os.path.split(
output_file_name)
extensionless_output_file_name, output_file_extension = os.path.splitext(
pathless_output_file_name)
if randomize_weights:
conv_dense_layer_names = _find_conv_and_dense_layers(model_object)
conv_dense_layer_names.reverse()
num_sets = len(conv_dense_layer_names)
else:
conv_dense_layer_names = []
num_sets = 1
print(
'Reading storm metadata from: "{0:s}"...'.format(storm_metafile_name))
full_storm_id_strings, storm_times_unix_sec = (
tracking_io.read_ids_and_times(storm_metafile_name))
print(SEPARATOR_STRING)
if 0 < num_examples < len(full_storm_id_strings):
full_storm_id_strings = full_storm_id_strings[:num_examples]
storm_times_unix_sec = storm_times_unix_sec[:num_examples]
example_dict = testing_io.read_predictors_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_full_id_strings=full_storm_id_strings,
desired_times_unix_sec=storm_times_unix_sec,
option_dict=training_option_dict,
layer_operation_dicts=model_metadata_dict[cnn.LAYER_OPERATIONS_KEY])
print(SEPARATOR_STRING)
predictor_matrices = example_dict[testing_io.INPUT_MATRICES_KEY]
sounding_pressure_matrix_pa = (
example_dict[testing_io.SOUNDING_PRESSURES_KEY])
print('Denormalizing model inputs...')
denorm_predictor_matrices = trainval_io.separate_shear_and_reflectivity(
list_of_input_matrices=copy.deepcopy(predictor_matrices),
training_option_dict=training_option_dict)
denorm_predictor_matrices = model_interpretation.denormalize_data(
list_of_input_matrices=denorm_predictor_matrices,
model_metadata_dict=model_metadata_dict)
print(SEPARATOR_STRING)
for k in range(num_sets):
if randomize_weights:
if cascading_random:
_reset_weights_in_layer(model_object=model_object,
layer_name=conv_dense_layer_names[k])
this_model_object = model_object
this_output_file_name = (
'{0:s}/{1:s}_cascading-random_{2:s}{3:s}').format(
output_dir_name, extensionless_output_file_name,
conv_dense_layer_names[k].replace('_', '-'),
output_file_extension)
else:
this_model_object = keras.models.Model.from_config(
model_object.get_config())
this_model_object.set_weights(model_object.get_weights())
_reset_weights_in_layer(model_object=this_model_object,
layer_name=conv_dense_layer_names[k])
this_output_file_name = '{0:s}/{1:s}_random_{2:s}{3:s}'.format(
output_dir_name, extensionless_output_file_name,
conv_dense_layer_names[k].replace('_', '-'),
output_file_extension)
else:
this_model_object = model_object
this_output_file_name = output_file_name
# print(K.eval(this_model_object.get_layer(name='dense_53').weights[0]))
these_cam_matrices, these_guided_cam_matrices = (
_run_gradcam_one_weight_set(
model_object=this_model_object,
target_class=target_class,
target_layer_name=target_layer_name,
predictor_matrices=predictor_matrices,
training_option_dict=training_option_dict))
print('Writing results to file: "{0:s}"...'.format(
this_output_file_name))
gradcam.write_standard_file(
pickle_file_name=this_output_file_name,
denorm_predictor_matrices=denorm_predictor_matrices,
cam_matrices=these_cam_matrices,
guided_cam_matrices=these_guided_cam_matrices,
full_storm_id_strings=full_storm_id_strings,
storm_times_unix_sec=storm_times_unix_sec,
model_file_name=model_file_name,
target_class=target_class,
target_layer_name=target_layer_name,
sounding_pressure_matrix_pa=sounding_pressure_matrix_pa)
print(SEPARATOR_STRING)
def _run(model_file_name, top_example_dir_name, first_spc_date_string,
last_spc_date_string, num_examples, class_fraction_keys,
class_fraction_values, num_bootstrap_iters,
bootstrap_confidence_level, output_file_name):
"""Runs permutation test for predictor importance.
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param top_example_dir_name: Same.
:param first_spc_date_string: Same.
:param last_spc_date_string: Same.
:param num_examples: Same.
:param class_fraction_keys: Same.
:param class_fraction_values: Same.
:param num_bootstrap_iters: Same.
:param bootstrap_confidence_level: Same.
:param output_file_name: Same.
"""
print('Reading model from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
model_directory_name, _ = os.path.split(model_file_name)
metadata_file_name = '{0:s}/model_metadata.p'.format(model_directory_name)
print('Reading metadata from: "{0:s}"...'.format(metadata_file_name))
model_metadata_dict = cnn.read_model_metadata(metadata_file_name)
training_option_dict = model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
if len(class_fraction_keys) > 1:
class_to_sampling_fraction_dict = dict(
list(zip(class_fraction_keys, class_fraction_values)))
else:
class_to_sampling_fraction_dict = None
training_option_dict[
trainval_io.SAMPLING_FRACTIONS_KEY] = class_to_sampling_fraction_dict
example_file_names = input_examples.find_many_example_files(
top_directory_name=top_example_dir_name,
shuffled=False,
first_spc_date_string=first_spc_date_string,
last_spc_date_string=last_spc_date_string,
raise_error_if_any_missing=False)
training_option_dict[trainval_io.EXAMPLE_FILES_KEY] = example_file_names
training_option_dict[trainval_io.FIRST_STORM_TIME_KEY] = (
time_conversion.get_start_of_spc_date(first_spc_date_string))
training_option_dict[trainval_io.LAST_STORM_TIME_KEY] = (
time_conversion.get_end_of_spc_date(last_spc_date_string))
if model_metadata_dict[cnn.LAYER_OPERATIONS_KEY] is not None:
generator_object = testing_io.gridrad_generator_2d_reduced(
option_dict=training_option_dict,
list_of_operation_dicts=model_metadata_dict[
cnn.LAYER_OPERATIONS_KEY],
num_examples_total=num_examples)
elif model_metadata_dict[cnn.CONV_2D3D_KEY]:
generator_object = testing_io.myrorss_generator_2d3d(
option_dict=training_option_dict, num_examples_total=num_examples)
else:
generator_object = testing_io.generator_2d_or_3d(
option_dict=training_option_dict, num_examples_total=num_examples)
full_id_strings = []
storm_times_unix_sec = numpy.array([], dtype=int)
target_values = numpy.array([], dtype=int)
list_of_predictor_matrices = None
print(SEPARATOR_STRING)
for _ in range(len(example_file_names)):
try:
this_storm_object_dict = next(generator_object)
print(SEPARATOR_STRING)
except StopIteration:
break
full_id_strings += this_storm_object_dict[testing_io.FULL_IDS_KEY]
storm_times_unix_sec = numpy.concatenate(
(storm_times_unix_sec,
this_storm_object_dict[testing_io.STORM_TIMES_KEY]))
these_target_values = this_storm_object_dict[
testing_io.TARGET_ARRAY_KEY]
if len(these_target_values.shape) > 1:
these_target_values = numpy.argmax(these_target_values, axis=1)
target_values = numpy.concatenate((target_values, these_target_values))
these_predictor_matrices = this_storm_object_dict[
testing_io.INPUT_MATRICES_KEY]
if list_of_predictor_matrices is None:
list_of_predictor_matrices = copy.deepcopy(
these_predictor_matrices)
else:
for k in range(len(list_of_predictor_matrices)):
list_of_predictor_matrices[k] = numpy.concatenate(
(list_of_predictor_matrices[k],
these_predictor_matrices[k]))
predictor_names_by_matrix = _create_predictor_names(
model_metadata_dict=model_metadata_dict,
list_of_predictor_matrices=list_of_predictor_matrices)
for i in range(len(predictor_names_by_matrix)):
print('Predictors in {0:d}th matrix:\n{1:s}\n'.format(
i + 1, str(predictor_names_by_matrix[i])))
print(SEPARATOR_STRING)
list_of_layer_operation_dicts = model_metadata_dict[
cnn.LAYER_OPERATIONS_KEY]
if list_of_layer_operation_dicts is not None:
correlation_matrix, predictor_names = _get_pearson_correlations(
list_of_predictor_matrices=list_of_predictor_matrices,
predictor_names_by_matrix=predictor_names_by_matrix,
sounding_heights_m_agl=training_option_dict[
trainval_io.SOUNDING_HEIGHTS_KEY])
for i in range(len(predictor_names)):
for j in range(i, len(predictor_names)):
print((
'Pearson correlation between "{0:s}" and "{1:s}" = {2:.4f}'
).format(predictor_names[i], predictor_names[j],
correlation_matrix[i, j]))
print('\n')
if model_metadata_dict[cnn.CONV_2D3D_KEY]:
prediction_function = permutation.prediction_function_2d3d_cnn
else:
num_radar_dimensions = len(list_of_predictor_matrices[0].shape) - 2
if num_radar_dimensions == 2:
prediction_function = permutation.prediction_function_2d_cnn
else:
prediction_function = permutation.prediction_function_3d_cnn
print(SEPARATOR_STRING)
result_dict = permutation.run_permutation_test(
model_object=model_object,
list_of_input_matrices=list_of_predictor_matrices,
predictor_names_by_matrix=predictor_names_by_matrix,
target_values=target_values,
prediction_function=prediction_function,
cost_function=permutation.negative_auc_function,
num_bootstrap_iters=num_bootstrap_iters,
bootstrap_confidence_level=bootstrap_confidence_level)
print(SEPARATOR_STRING)
result_dict[permutation.MODEL_FILE_KEY] = model_file_name
result_dict[permutation.TARGET_VALUES_KEY] = target_values
result_dict[permutation.FULL_IDS_KEY] = full_id_strings
result_dict[permutation.STORM_TIMES_KEY] = storm_times_unix_sec
print('Writing results to: "{0:s}"...'.format(output_file_name))
permutation.write_results(result_dict=result_dict,
pickle_file_name=output_file_name)
def _run(input_model_file_name, sounding_field_names,
normalization_type_string, normalization_param_file_name,
min_normalized_value, max_normalized_value, target_name,
downsampling_classes, downsampling_fractions, monitor_string,
weight_loss_function, x_translations_pixels, y_translations_pixels,
ccw_rotation_angles_deg, noise_standard_deviation, num_noisings,
flip_in_x, flip_in_y, top_training_dir_name,
first_training_time_string, last_training_time_string,
num_examples_per_train_batch, top_validation_dir_name,
first_validation_time_string, last_validation_time_string,
num_examples_per_validn_batch, num_epochs,
num_training_batches_per_epoch, num_validation_batches_per_epoch,
output_dir_name):
"""Trains CNN with 2-D and 3-D MYRORSS images.
This is effectively the main method.
:param input_model_file_name: See documentation at top of file.
:param sounding_field_names: Same.
:param normalization_type_string: Same.
:param normalization_param_file_name: Same.
:param min_normalized_value: Same.
:param max_normalized_value: Same.
:param target_name: Same.
:param downsampling_classes: Same.
:param downsampling_fractions: Same.
:param monitor_string: Same.
:param weight_loss_function: Same.
:param x_translations_pixels: Same.
:param y_translations_pixels: Same.
:param ccw_rotation_angles_deg: Same.
:param noise_standard_deviation: Same.
:param num_noisings: Same.
:param flip_in_x: Same.
:param flip_in_y: Same.
:param top_training_dir_name: Same.
:param first_training_time_string: Same.
:param last_training_time_string: Same.
:param num_examples_per_train_batch: Same.
:param top_validation_dir_name: Same.
:param first_validation_time_string: Same.
:param last_validation_time_string: Same.
:param num_examples_per_validn_batch: Same.
:param num_epochs: Same.
:param num_training_batches_per_epoch: Same.
:param num_validation_batches_per_epoch: Same.
:param output_dir_name: Same.
"""
file_system_utils.mkdir_recursive_if_necessary(
directory_name=output_dir_name)
# argument_file_name = '{0:s}/input_args.p'.format(output_dir_name)
# print('Writing input args to: "{0:s}"...'.format(argument_file_name))
#
# argument_file_handle = open(argument_file_name, 'wb')
# pickle.dump(INPUT_ARG_OBJECT.__dict__, argument_file_handle)
# argument_file_handle.close()
#
# return
# Process input args.
first_training_time_unix_sec = time_conversion.string_to_unix_sec(
first_training_time_string, TIME_FORMAT)
last_training_time_unix_sec = time_conversion.string_to_unix_sec(
last_training_time_string, TIME_FORMAT)
first_validation_time_unix_sec = time_conversion.string_to_unix_sec(
first_validation_time_string, TIME_FORMAT)
last_validation_time_unix_sec = time_conversion.string_to_unix_sec(
last_validation_time_string, TIME_FORMAT)
if sounding_field_names[0] in ['', 'None']:
sounding_field_names = None
if len(downsampling_classes) > 1:
downsampling_dict = dict(
list(zip(downsampling_classes, downsampling_fractions)))
else:
downsampling_dict = None
if (len(x_translations_pixels) == 1
and x_translations_pixels + y_translations_pixels == 0):
x_translations_pixels = None
y_translations_pixels = None
if len(ccw_rotation_angles_deg) == 1 and ccw_rotation_angles_deg[0] == 0:
ccw_rotation_angles_deg = None
if num_noisings <= 0:
num_noisings = 0
noise_standard_deviation = None
# Set output locations.
output_model_file_name = '{0:s}/model.h5'.format(output_dir_name)
history_file_name = '{0:s}/model_history.csv'.format(output_dir_name)
tensorboard_dir_name = '{0:s}/tensorboard'.format(output_dir_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(output_dir_name)
# Find training and validation files.
training_file_names = input_examples.find_many_example_files(
top_directory_name=top_training_dir_name,
shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER,
raise_error_if_any_missing=False)
validation_file_names = input_examples.find_many_example_files(
top_directory_name=top_validation_dir_name,
shuffled=True,
first_batch_number=FIRST_BATCH_NUMBER,
last_batch_number=LAST_BATCH_NUMBER,
raise_error_if_any_missing=False)
# Read architecture.
print(
'Reading architecture from: "{0:s}"...'.format(input_model_file_name))
model_object = cnn.read_model(input_model_file_name)
# model_object = keras.models.clone_model(model_object)
# TODO(thunderhoser): This is a HACK.
model_object.compile(loss=keras.losses.binary_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=cnn_setup.DEFAULT_METRIC_FUNCTION_LIST)
print(SEPARATOR_STRING)
model_object.summary()
print(SEPARATOR_STRING)
# Write metadata.
metadata_dict = {
cnn.NUM_EPOCHS_KEY: num_epochs,
cnn.NUM_TRAINING_BATCHES_KEY: num_training_batches_per_epoch,
cnn.NUM_VALIDATION_BATCHES_KEY: num_validation_batches_per_epoch,
cnn.MONITOR_STRING_KEY: monitor_string,
cnn.WEIGHT_LOSS_FUNCTION_KEY: weight_loss_function,
cnn.CONV_2D3D_KEY: True,
cnn.VALIDATION_FILES_KEY: validation_file_names,
cnn.FIRST_VALIDN_TIME_KEY: first_validation_time_unix_sec,
cnn.LAST_VALIDN_TIME_KEY: last_validation_time_unix_sec,
cnn.NUM_EX_PER_VALIDN_BATCH_KEY: num_examples_per_validn_batch
}
if isinstance(model_object.input, list):
list_of_input_tensors = model_object.input
else:
list_of_input_tensors = [model_object.input]
upsample_refl = len(list_of_input_tensors) == 2
num_grid_rows = list_of_input_tensors[0].get_shape().as_list()[1]
num_grid_columns = list_of_input_tensors[0].get_shape().as_list()[2]
if upsample_refl:
num_grid_rows = int(numpy.round(num_grid_rows / 2))
num_grid_columns = int(numpy.round(num_grid_columns / 2))
training_option_dict = {
trainval_io.EXAMPLE_FILES_KEY: training_file_names,
trainval_io.TARGET_NAME_KEY: target_name,
trainval_io.FIRST_STORM_TIME_KEY: first_training_time_unix_sec,
trainval_io.LAST_STORM_TIME_KEY: last_training_time_unix_sec,
trainval_io.NUM_EXAMPLES_PER_BATCH_KEY: num_examples_per_train_batch,
trainval_io.RADAR_FIELDS_KEY:
input_examples.AZIMUTHAL_SHEAR_FIELD_NAMES,
trainval_io.RADAR_HEIGHTS_KEY: REFLECTIVITY_HEIGHTS_M_AGL,
trainval_io.SOUNDING_FIELDS_KEY: sounding_field_names,
trainval_io.SOUNDING_HEIGHTS_KEY: SOUNDING_HEIGHTS_M_AGL,
trainval_io.NUM_ROWS_KEY: num_grid_rows,
trainval_io.NUM_COLUMNS_KEY: num_grid_columns,
trainval_io.NORMALIZATION_TYPE_KEY: normalization_type_string,
trainval_io.NORMALIZATION_FILE_KEY: normalization_param_file_name,
trainval_io.MIN_NORMALIZED_VALUE_KEY: min_normalized_value,
trainval_io.MAX_NORMALIZED_VALUE_KEY: max_normalized_value,
trainval_io.BINARIZE_TARGET_KEY: False,
trainval_io.SAMPLING_FRACTIONS_KEY: downsampling_dict,
trainval_io.LOOP_ONCE_KEY: False,
trainval_io.X_TRANSLATIONS_KEY: x_translations_pixels,
trainval_io.Y_TRANSLATIONS_KEY: y_translations_pixels,
trainval_io.ROTATION_ANGLES_KEY: ccw_rotation_angles_deg,
trainval_io.NOISE_STDEV_KEY: noise_standard_deviation,
trainval_io.NUM_NOISINGS_KEY: num_noisings,
trainval_io.FLIP_X_KEY: flip_in_x,
trainval_io.FLIP_Y_KEY: flip_in_y,
trainval_io.UPSAMPLE_REFLECTIVITY_KEY: upsample_refl
}
print('Writing metadata to: "{0:s}"...'.format(model_metafile_name))
cnn.write_model_metadata(pickle_file_name=model_metafile_name,
metadata_dict=metadata_dict,
training_option_dict=training_option_dict)
cnn.train_cnn_2d3d_myrorss(
model_object=model_object,
model_file_name=output_model_file_name,
history_file_name=history_file_name,
tensorboard_dir_name=tensorboard_dir_name,
num_epochs=num_epochs,
num_training_batches_per_epoch=num_training_batches_per_epoch,
training_option_dict=training_option_dict,
monitor_string=monitor_string,
weight_loss_function=weight_loss_function,
num_validation_batches_per_epoch=num_validation_batches_per_epoch,
validation_file_names=validation_file_names,
first_validn_time_unix_sec=first_validation_time_unix_sec,
last_validn_time_unix_sec=last_validation_time_unix_sec,
num_examples_per_validn_batch=num_examples_per_validn_batch)
def _run(model_file_name, component_type_string, target_class, layer_name,
ideal_activation, neuron_indices, channel_index, top_example_dir_name,
storm_metafile_name, num_examples, randomize_weights,
cascading_random, output_file_name):
"""Computes saliency map for each storm object and each model component.
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param component_type_string: Same.
:param target_class: Same.
:param layer_name: Same.
:param ideal_activation: Same.
:param neuron_indices: Same.
:param channel_index: Same.
:param top_example_dir_name: Same.
:param storm_metafile_name: Same.
:param num_examples: Same.
:param randomize_weights: Same.
:param cascading_random: Same.
:param output_file_name: Same.
"""
# Check input args.
file_system_utils.mkdir_recursive_if_necessary(file_name=output_file_name)
model_interpretation.check_component_type(component_type_string)
# Read model and metadata.
print('Reading model from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(model_file_name)[0])
print(
'Reading model metadata from: "{0:s}"...'.format(model_metafile_name))
model_metadata_dict = cnn.read_model_metadata(model_metafile_name)
training_option_dict = model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
training_option_dict[trainval_io.REFLECTIVITY_MASK_KEY] = None
output_dir_name, pathless_output_file_name = os.path.split(
output_file_name)
extensionless_output_file_name, output_file_extension = os.path.splitext(
pathless_output_file_name)
if randomize_weights:
conv_dense_layer_names = _find_conv_and_dense_layers(model_object)
conv_dense_layer_names.reverse()
num_sets = len(conv_dense_layer_names)
else:
conv_dense_layer_names = []
num_sets = 1
print(
'Reading storm metadata from: "{0:s}"...'.format(storm_metafile_name))
full_storm_id_strings, storm_times_unix_sec = (
tracking_io.read_ids_and_times(storm_metafile_name))
print(SEPARATOR_STRING)
if 0 < num_examples < len(full_storm_id_strings):
full_storm_id_strings = full_storm_id_strings[:num_examples]
storm_times_unix_sec = storm_times_unix_sec[:num_examples]
example_dict = testing_io.read_predictors_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_full_id_strings=full_storm_id_strings,
desired_times_unix_sec=storm_times_unix_sec,
option_dict=training_option_dict,
layer_operation_dicts=model_metadata_dict[cnn.LAYER_OPERATIONS_KEY])
print(SEPARATOR_STRING)
predictor_matrices = example_dict[testing_io.INPUT_MATRICES_KEY]
sounding_pressure_matrix_pa = example_dict[
testing_io.SOUNDING_PRESSURES_KEY]
denorm_predictor_matrices = trainval_io.separate_shear_and_reflectivity(
list_of_input_matrices=copy.deepcopy(predictor_matrices),
training_option_dict=training_option_dict)
print('Denormalizing model inputs...')
denorm_predictor_matrices = model_interpretation.denormalize_data(
list_of_input_matrices=denorm_predictor_matrices,
model_metadata_dict=model_metadata_dict)
print(SEPARATOR_STRING)
for k in range(num_sets):
if randomize_weights:
if cascading_random:
_reset_weights_in_layer(model_object=model_object,
layer_name=conv_dense_layer_names[k])
this_model_object = model_object
this_output_file_name = (
'{0:s}/{1:s}_cascading-random_{2:s}{3:s}').format(
output_dir_name, extensionless_output_file_name,
conv_dense_layer_names[k].replace('_', '-'),
output_file_extension)
else:
this_model_object = keras.models.Model.from_config(
model_object.get_config())
this_model_object.set_weights(model_object.get_weights())
_reset_weights_in_layer(model_object=this_model_object,
layer_name=conv_dense_layer_names[k])
this_output_file_name = '{0:s}/{1:s}_random_{2:s}{3:s}'.format(
output_dir_name, extensionless_output_file_name,
conv_dense_layer_names[k].replace('_', '-'),
output_file_extension)
else:
this_model_object = model_object
this_output_file_name = output_file_name
# print(K.eval(this_model_object.get_layer(name='dense_3').weights[0]))
if component_type_string == CLASS_COMPONENT_TYPE_STRING:
print('Computing saliency maps for target class {0:d}...'.format(
target_class))
saliency_matrices = (
saliency_maps.get_saliency_maps_for_class_activation(
model_object=this_model_object,
target_class=target_class,
list_of_input_matrices=predictor_matrices))
elif component_type_string == NEURON_COMPONENT_TYPE_STRING:
print(
('Computing saliency maps for neuron {0:s} in layer "{1:s}"...'
).format(str(neuron_indices), layer_name))
saliency_matrices = (
saliency_maps.get_saliency_maps_for_neuron_activation(
model_object=this_model_object,
layer_name=layer_name,
neuron_indices=neuron_indices,
list_of_input_matrices=predictor_matrices,
ideal_activation=ideal_activation))
else:
print((
'Computing saliency maps for channel {0:d} in layer "{1:s}"...'
).format(channel_index, layer_name))
saliency_matrices = (
saliency_maps.get_saliency_maps_for_channel_activation(
model_object=this_model_object,
layer_name=layer_name,
channel_index=channel_index,
list_of_input_matrices=predictor_matrices,
stat_function_for_neuron_activations=K.max,
ideal_activation=ideal_activation))
saliency_matrices = trainval_io.separate_shear_and_reflectivity(
list_of_input_matrices=saliency_matrices,
training_option_dict=training_option_dict)
print('Writing saliency maps to file: "{0:s}"...'.format(
this_output_file_name))
saliency_metadata_dict = saliency_maps.check_metadata(
component_type_string=component_type_string,
target_class=target_class,
layer_name=layer_name,
ideal_activation=ideal_activation,
neuron_indices=neuron_indices,
channel_index=channel_index)
saliency_maps.write_standard_file(
pickle_file_name=this_output_file_name,
denorm_predictor_matrices=denorm_predictor_matrices,
saliency_matrices=saliency_matrices,
full_storm_id_strings=full_storm_id_strings,
storm_times_unix_sec=storm_times_unix_sec,
model_file_name=model_file_name,
metadata_dict=saliency_metadata_dict,
sounding_pressure_matrix_pa=sounding_pressure_matrix_pa)
def _train_one_cnn(gpu_queue, argument_dict):
"""Trains single CNN with 2-D GridRad data.
:param gpu_queue: GPU queue (instance of `multiprocessing.Manager.Queue`).
:param argument_dict: Dictionary of CNN arguments, where each key is an
input arg to the script train_cnn_gridrad_2d_reduced.py.
"""
import keras
from keras import backend as K
import tensorflow
from gewittergefahr.deep_learning import cnn
from gewittergefahr.deep_learning import cnn_setup
from gewittergefahr.scripts import deep_learning_helper as dl_helper
gpu_index = -1
try:
# Deal with GPU business.
gpu_index = int(gpu_queue.get())
os.environ['CUDA_VISIBLE_DEVICES'] = '{0:d}'.format(gpu_index)
session_object = tensorflow.Session(
config=tensorflow.ConfigProto(intra_op_parallelism_threads=7,
inter_op_parallelism_threads=7,
allow_soft_placement=False,
log_device_placement=False,
gpu_options=tensorflow.GPUOptions(
allow_growth=True)))
K.set_session(session_object)
# Read untrained model.
untrained_model_file_name = argument_dict[
dl_helper.INPUT_MODEL_FILE_ARG_NAME]
with tensorflow.device('/gpu:0'):
print('Reading untrained model from: "{0:s}"...'.format(
untrained_model_file_name))
model_object = cnn.read_model(untrained_model_file_name)
model_object.compile(loss=keras.losses.binary_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=cnn_setup.DEFAULT_METRIC_FUNCTION_LIST)
print(SEPARATOR_STRING)
model_object.summary()
print(SEPARATOR_STRING)
# Write metadata.
metadata_dict, training_option_dict = _write_metadata_one_cnn(
model_object=model_object, argument_dict=argument_dict)
print('Training CNN on GPU {0:d}...'.format(gpu_index))
print(SEPARATOR_STRING)
# Train CNN.
output_dir_name = argument_dict[dl_helper.OUTPUT_DIR_ARG_NAME]
output_model_file_name = '{0:s}/model.h5'.format(output_dir_name)
history_file_name = '{0:s}/model_history.csv'.format(output_dir_name)
tensorboard_dir_name = '{0:s}/tensorboard'.format(output_dir_name)
cnn.train_cnn_gridrad_2d_reduced(
model_object=model_object,
model_file_name=output_model_file_name,
history_file_name=history_file_name,
tensorboard_dir_name=tensorboard_dir_name,
num_epochs=metadata_dict[cnn.NUM_EPOCHS_KEY],
num_training_batches_per_epoch=metadata_dict[
cnn.NUM_TRAINING_BATCHES_KEY],
training_option_dict=training_option_dict,
list_of_layer_operation_dicts=metadata_dict[
cnn.LAYER_OPERATIONS_KEY],
monitor_string=metadata_dict[cnn.MONITOR_STRING_KEY],
weight_loss_function=metadata_dict[cnn.WEIGHT_LOSS_FUNCTION_KEY],
num_validation_batches_per_epoch=metadata_dict[
cnn.NUM_VALIDATION_BATCHES_KEY],
validation_file_names=metadata_dict[cnn.VALIDATION_FILES_KEY],
first_validn_time_unix_sec=metadata_dict[
cnn.FIRST_VALIDN_TIME_KEY],
last_validn_time_unix_sec=metadata_dict[cnn.LAST_VALIDN_TIME_KEY],
num_examples_per_validn_batch=metadata_dict[
cnn.NUM_EX_PER_VALIDN_BATCH_KEY])
session_object.close()
del session_object
gpu_queue.put(gpu_index)
except Exception as this_exception:
if gpu_index >= 0:
gpu_queue.put(gpu_index)
print(traceback.format_exc())
raise this_exception
def _run(model_file_name, component_type_string, target_class, layer_name,
neuron_indices_flattened, channel_indices, top_example_dir_name,
first_spc_date_string, last_spc_date_string, output_file_name):
"""Creates activation maps for one class, neuron, or channel of a CNN.
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param component_type_string: Same.
:param target_class: Same.
:param layer_name: Same.
:param neuron_indices_flattened: Same.
:param channel_indices: Same.
:param top_example_dir_name: Same.
:param first_spc_date_string: Same.
:param last_spc_date_string: Same.
:param output_file_name: Same.
"""
# Check input args.
file_system_utils.mkdir_recursive_if_necessary(file_name=output_file_name)
model_interpretation.check_component_type(component_type_string)
if component_type_string == CHANNEL_COMPONENT_TYPE_STRING:
error_checking.assert_is_geq_numpy_array(channel_indices, 0)
if component_type_string == NEURON_COMPONENT_TYPE_STRING:
neuron_indices_flattened = neuron_indices_flattened.astype(float)
neuron_indices_flattened[neuron_indices_flattened < 0] = numpy.nan
neuron_indices_2d_list = general_utils.split_array_by_nan(
neuron_indices_flattened)
neuron_index_matrix = numpy.array(neuron_indices_2d_list, dtype=int)
else:
neuron_index_matrix = None
# Read model and metadata.
print('Reading model from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
metadata_file_name = '{0:s}/model_metadata.p'.format(
os.path.split(model_file_name)[0])
print('Reading metadata from: "{0:s}"...'.format(metadata_file_name))
model_metadata_dict = cnn.read_model_metadata(metadata_file_name)
training_option_dict = model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
# Create generator.
example_file_names = input_examples.find_many_example_files(
top_directory_name=top_example_dir_name,
shuffled=False,
first_spc_date_string=first_spc_date_string,
last_spc_date_string=last_spc_date_string,
raise_error_if_any_missing=False)
training_option_dict[trainval_io.SAMPLING_FRACTIONS_KEY] = None
training_option_dict[trainval_io.EXAMPLE_FILES_KEY] = example_file_names
training_option_dict[trainval_io.FIRST_STORM_TIME_KEY] = (
time_conversion.get_start_of_spc_date(first_spc_date_string))
training_option_dict[trainval_io.LAST_STORM_TIME_KEY] = (
time_conversion.get_end_of_spc_date(last_spc_date_string))
if model_metadata_dict[cnn.LAYER_OPERATIONS_KEY] is not None:
generator_object = testing_io.gridrad_generator_2d_reduced(
option_dict=training_option_dict,
list_of_operation_dicts=model_metadata_dict[
cnn.LAYER_OPERATIONS_KEY],
num_examples_total=LARGE_INTEGER)
elif model_metadata_dict[cnn.CONV_2D3D_KEY]:
generator_object = testing_io.myrorss_generator_2d3d(
option_dict=training_option_dict, num_examples_total=LARGE_INTEGER)
else:
generator_object = testing_io.generator_2d_or_3d(
option_dict=training_option_dict, num_examples_total=LARGE_INTEGER)
# Compute activation for each example (storm object) and model component.
full_id_strings = []
storm_times_unix_sec = numpy.array([], dtype=int)
activation_matrix = None
print(SEPARATOR_STRING)
for _ in range(len(example_file_names)):
try:
this_storm_object_dict = next(generator_object)
except StopIteration:
break
this_list_of_input_matrices = this_storm_object_dict[
testing_io.INPUT_MATRICES_KEY]
these_id_strings = this_storm_object_dict[testing_io.FULL_IDS_KEY]
these_times_unix_sec = this_storm_object_dict[
testing_io.STORM_TIMES_KEY]
full_id_strings += these_id_strings
storm_times_unix_sec = numpy.concatenate(
(storm_times_unix_sec, these_times_unix_sec))
if component_type_string == CLASS_COMPONENT_TYPE_STRING:
print('Computing activations for target class {0:d}...'.format(
target_class))
this_activation_matrix = (
model_activation.get_class_activation_for_examples(
model_object=model_object,
target_class=target_class,
list_of_input_matrices=this_list_of_input_matrices))
this_activation_matrix = numpy.reshape(
this_activation_matrix, (len(this_activation_matrix), 1))
elif component_type_string == NEURON_COMPONENT_TYPE_STRING:
this_activation_matrix = None
for j in range(neuron_index_matrix.shape[0]):
print((
'Computing activations for neuron {0:s} in layer "{1:s}"...'
).format(str(neuron_index_matrix[j, :]), layer_name))
these_activations = (
model_activation.get_neuron_activation_for_examples(
model_object=model_object,
layer_name=layer_name,
neuron_indices=neuron_index_matrix[j, :],
list_of_input_matrices=this_list_of_input_matrices))
these_activations = numpy.reshape(these_activations,
(len(these_activations), 1))
if this_activation_matrix is None:
this_activation_matrix = these_activations + 0.
else:
this_activation_matrix = numpy.concatenate(
(this_activation_matrix, these_activations), axis=1)
else:
this_activation_matrix = None
for this_channel_index in channel_indices:
print(('Computing activations for channel {0:d} in layer '
'"{1:s}"...').format(this_channel_index, layer_name))
these_activations = (
model_activation.get_channel_activation_for_examples(
model_object=model_object,
layer_name=layer_name,
channel_index=this_channel_index,
list_of_input_matrices=this_list_of_input_matrices,
stat_function_for_neuron_activations=K.max))
these_activations = numpy.reshape(these_activations,
(len(these_activations), 1))
if this_activation_matrix is None:
this_activation_matrix = these_activations + 0.
else:
this_activation_matrix = numpy.concatenate(
(this_activation_matrix, these_activations), axis=1)
if activation_matrix is None:
activation_matrix = this_activation_matrix + 0.
else:
activation_matrix = numpy.concatenate(
(activation_matrix, this_activation_matrix), axis=0)
print(SEPARATOR_STRING)
print('Writing activations to file: "{0:s}"...'.format(output_file_name))
model_activation.write_file(pickle_file_name=output_file_name,
activation_matrix=activation_matrix,
full_id_strings=full_id_strings,
storm_times_unix_sec=storm_times_unix_sec,
model_file_name=model_file_name,
component_type_string=component_type_string,
target_class=target_class,
layer_name=layer_name,
neuron_index_matrix=neuron_index_matrix,
channel_indices=channel_indices)
def _run(model_file_name, layer_names, top_example_dir_name,
storm_metafile_name, num_examples, top_output_dir_name):
"""Evaluates CNN (convolutional neural net) predictions.
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param layer_names: Same.
:param top_example_dir_name: Same.
:param storm_metafile_name: Same.
:param num_examples: Same.
:param top_output_dir_name: Same.
:raises: ValueError: if feature maps do not have 2 or 3 spatial dimensions.
"""
print('Reading model from: "{0:s}"...'.format(model_file_name))
model_object = cnn.read_model(model_file_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(model_file_name)[0])
print(
'Reading model metadata from: "{0:s}"...'.format(model_metafile_name))
model_metadata_dict = cnn.read_model_metadata(model_metafile_name)
training_option_dict = model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
training_option_dict[trainval_io.REFLECTIVITY_MASK_KEY] = None
print(
'Reading storm metadata from: "{0:s}"...'.format(storm_metafile_name))
full_id_strings, storm_times_unix_sec = tracking_io.read_ids_and_times(
storm_metafile_name)
print(SEPARATOR_STRING)
if 0 < num_examples < len(full_id_strings):
full_id_strings = full_id_strings[:num_examples]
storm_times_unix_sec = storm_times_unix_sec[:num_examples]
list_of_predictor_matrices = testing_io.read_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_full_id_strings=full_id_strings,
desired_times_unix_sec=storm_times_unix_sec,
option_dict=training_option_dict,
list_of_layer_operation_dicts=model_metadata_dict[
cnn.LAYER_OPERATIONS_KEY])[0]
print(SEPARATOR_STRING)
include_soundings = (training_option_dict[trainval_io.SOUNDING_FIELDS_KEY]
is not None)
if include_soundings:
sounding_matrix = list_of_predictor_matrices[-1]
else:
sounding_matrix = None
num_layers = len(layer_names)
feature_matrix_by_layer = [None] * num_layers
for k in range(num_layers):
if model_metadata_dict[cnn.CONV_2D3D_KEY]:
if training_option_dict[trainval_io.UPSAMPLE_REFLECTIVITY_KEY]:
feature_matrix_by_layer[k] = cnn.apply_2d_or_3d_cnn(
model_object=model_object,
radar_image_matrix=list_of_predictor_matrices[0],
sounding_matrix=sounding_matrix,
return_features=True,
feature_layer_name=layer_names[k])
else:
feature_matrix_by_layer[k] = cnn.apply_2d3d_cnn(
model_object=model_object,
reflectivity_matrix_dbz=list_of_predictor_matrices[0],
azimuthal_shear_matrix_s01=list_of_predictor_matrices[1],
sounding_matrix=sounding_matrix,
return_features=True,
feature_layer_name=layer_names[k])
else:
feature_matrix_by_layer[k] = cnn.apply_2d_or_3d_cnn(
model_object=model_object,
radar_image_matrix=list_of_predictor_matrices[0],
sounding_matrix=sounding_matrix,
return_features=True,
feature_layer_name=layer_names[k])
for k in range(num_layers):
this_output_dir_name = '{0:s}/{1:s}'.format(top_output_dir_name,
layer_names[k])
file_system_utils.mkdir_recursive_if_necessary(
directory_name=this_output_dir_name)
_plot_feature_maps_one_layer(feature_matrix=feature_matrix_by_layer[k],
full_id_strings=full_id_strings,
storm_times_unix_sec=storm_times_unix_sec,
layer_name=layer_names[k],
output_dir_name=this_output_dir_name)
print(SEPARATOR_STRING)
def _run(model_file_name, component_type_string, target_class, layer_name,
ideal_activation, neuron_indices, channel_index, top_example_dir_name,
storm_metafile_name, num_examples, output_file_name):
"""Computes saliency map for each storm object and each model component.
This is effectively the main method.
:param model_file_name: See documentation at top of file.
:param component_type_string: Same.
:param target_class: Same.
:param layer_name: Same.
:param ideal_activation: Same.
:param neuron_indices: Same.
:param channel_index: Same.
:param top_example_dir_name: Same.
:param storm_metafile_name: Same.
:param num_examples: Same.
:param output_file_name: Same.
"""
# Check input args.
file_system_utils.mkdir_recursive_if_necessary(file_name=output_file_name)
model_interpretation.check_component_type(component_type_string)
# Read model and metadata.
print 'Reading model from: "{0:s}"...'.format(model_file_name)
model_object = cnn.read_model(model_file_name)
model_metafile_name = '{0:s}/model_metadata.p'.format(
os.path.split(model_file_name)[0])
print 'Reading model metadata from: "{0:s}"...'.format(model_metafile_name)
model_metadata_dict = cnn.read_model_metadata(model_metafile_name)
training_option_dict = model_metadata_dict[cnn.TRAINING_OPTION_DICT_KEY]
training_option_dict[trainval_io.REFLECTIVITY_MASK_KEY] = None
print 'Reading storm metadata from: "{0:s}"...'.format(storm_metafile_name)
storm_ids, storm_times_unix_sec = tracking_io.read_ids_and_times(
storm_metafile_name)
print SEPARATOR_STRING
if 0 < num_examples < len(storm_ids):
storm_ids = storm_ids[:num_examples]
storm_times_unix_sec = storm_times_unix_sec[:num_examples]
list_of_input_matrices, sounding_pressure_matrix_pascals = (
testing_io.read_specific_examples(
top_example_dir_name=top_example_dir_name,
desired_storm_ids=storm_ids,
desired_times_unix_sec=storm_times_unix_sec,
option_dict=training_option_dict,
list_of_layer_operation_dicts=model_metadata_dict[
cnn.LAYER_OPERATIONS_KEY]))
print SEPARATOR_STRING
if component_type_string == CLASS_COMPONENT_TYPE_STRING:
print 'Computing saliency maps for target class {0:d}...'.format(
target_class)
list_of_saliency_matrices = (
saliency_maps.get_saliency_maps_for_class_activation(
model_object=model_object,
target_class=target_class,
list_of_input_matrices=list_of_input_matrices))
elif component_type_string == NEURON_COMPONENT_TYPE_STRING:
print('Computing saliency maps for neuron {0:s} in layer "{1:s}"...'
).format(str(neuron_indices), layer_name)
list_of_saliency_matrices = (
saliency_maps.get_saliency_maps_for_neuron_activation(
model_object=model_object,
layer_name=layer_name,
neuron_indices=neuron_indices,
list_of_input_matrices=list_of_input_matrices,
ideal_activation=ideal_activation))
else:
print('Computing saliency maps for channel {0:d} in layer "{1:s}"...'
).format(channel_index, layer_name)
list_of_saliency_matrices = (
saliency_maps.get_saliency_maps_for_channel_activation(
model_object=model_object,
layer_name=layer_name,
channel_index=channel_index,
list_of_input_matrices=list_of_input_matrices,
stat_function_for_neuron_activations=K.max,
ideal_activation=ideal_activation))
print 'Denormalizing model inputs...'
list_of_input_matrices = model_interpretation.denormalize_data(
list_of_input_matrices=list_of_input_matrices,
model_metadata_dict=model_metadata_dict)
print 'Writing saliency maps to file: "{0:s}"...'.format(output_file_name)
saliency_metadata_dict = saliency_maps.check_metadata(
component_type_string=component_type_string,
target_class=target_class,
layer_name=layer_name,
ideal_activation=ideal_activation,
neuron_indices=neuron_indices,
channel_index=channel_index)
saliency_maps.write_standard_file(
pickle_file_name=output_file_name,
list_of_input_matrices=list_of_input_matrices,
list_of_saliency_matrices=list_of_saliency_matrices,
storm_ids=storm_ids,
storm_times_unix_sec=storm_times_unix_sec,
model_file_name=model_file_name,
saliency_metadata_dict=saliency_metadata_dict,
sounding_pressure_matrix_pascals=sounding_pressure_matrix_pascals)
