def test_determinize_probabilities(self):
"""Ensures correct output from determinize_probabilities."""
this_label_matrix = object_based_eval.determinize_probabilities(
class_probability_matrix=PROBABILITY_MATRIX,
binarization_threshold=BINARIZATION_THRESHOLD)
self.assertTrue(
numpy.array_equal(this_label_matrix, PREDICTED_LABEL_MATRIX))
def _run():
"""Plots determinization of front probabilities.
This is effectively the main method.
"""
prediction_dict = ml_utils.read_gridded_predictions(PREDICTION_FILE_NAME)
class_probability_matrix = prediction_dict[ml_utils.PROBABILITY_MATRIX_KEY]
for this_id in front_utils.VALID_INTEGER_IDS:
if this_id == front_utils.NO_FRONT_INTEGER_ID:
class_probability_matrix[..., this_id][numpy.isnan(
class_probability_matrix[..., this_id])] = 1.
else:
class_probability_matrix[..., this_id][numpy.isnan(
class_probability_matrix[..., this_id])] = 0.
_plot_predictions(output_file_name=BEFORE_FILE_NAME,
title_string='Probabilities',
annotation_string='(a)',
class_probability_matrix=class_probability_matrix)
predicted_label_matrix = object_eval.determinize_probabilities(
class_probability_matrix=class_probability_matrix,
binarization_threshold=BINARIZATION_THRESHOLD)
_plot_predictions(output_file_name=AFTER_FILE_NAME,
title_string='Deterministic predictions',
annotation_string='(b)',
predicted_label_matrix=predicted_label_matrix)
print 'Concatenating figures to: "{0:s}"...'.format(CONCAT_FILE_NAME)
imagemagick_utils.concatenate_images(
input_file_names=[BEFORE_FILE_NAME, AFTER_FILE_NAME],
output_file_name=CONCAT_FILE_NAME,
num_panel_rows=2,
num_panel_columns=1)
imagemagick_utils.resize_image(input_file_name=CONCAT_FILE_NAME,
output_file_name=CONCAT_FILE_NAME,
output_size_pixels=CONCAT_SIZE_PIXELS)
def _run():
"""Plots conversion of gridded probabilities into objects.
This is effectively the main method.
"""
prediction_dict = ml_utils.read_gridded_predictions(PREDICTION_FILE_NAME)
class_probability_matrix = prediction_dict[ml_utils.PROBABILITY_MATRIX_KEY]
for this_id in front_utils.VALID_INTEGER_IDS:
if this_id == front_utils.NO_FRONT_INTEGER_ID:
class_probability_matrix[..., this_id][numpy.isnan(
class_probability_matrix[..., this_id])] = 1.
else:
class_probability_matrix[..., this_id][numpy.isnan(
class_probability_matrix[..., this_id])] = 0.
predicted_label_matrix = object_eval.determinize_probabilities(
class_probability_matrix=class_probability_matrix,
binarization_threshold=BINARIZATION_THRESHOLD)
_plot_predictions(predicted_label_matrix=predicted_label_matrix,
title_string='All frontal regions',
annotation_string='(b)',
output_file_name=ALL_REGIONS_FILE_NAME)
valid_time_unix_sec = time_conversion.string_to_unix_sec(
VALID_TIME_STRING, TIME_FORMAT)
valid_times_unix_sec = numpy.array([valid_time_unix_sec], dtype=int)
predicted_region_table = object_eval.images_to_regions(
predicted_label_matrix=predicted_label_matrix,
image_times_unix_sec=valid_times_unix_sec)
num_grid_rows, num_grid_columns = nwp_model_utils.get_grid_dimensions(
model_name=nwp_model_utils.NARR_MODEL_NAME)
grid_spacing_metres = nwp_model_utils.get_xy_grid_spacing(
model_name=nwp_model_utils.NARR_MODEL_NAME)[0]
predicted_region_table = object_eval.discard_regions_with_small_area(
predicted_region_table=predicted_region_table,
x_grid_spacing_metres=grid_spacing_metres,
y_grid_spacing_metres=grid_spacing_metres,
min_area_metres2=MIN_REGION_AREA_METRES2)
predicted_label_matrix = object_eval.regions_to_images(
predicted_region_table=predicted_region_table,
num_grid_rows=num_grid_rows,
num_grid_columns=num_grid_columns)
_plot_predictions(predicted_label_matrix=predicted_label_matrix,
title_string='Large frontal regions',
annotation_string='(c)',
output_file_name=LARGE_REGIONS_FILE_NAME)
predicted_region_table = object_eval.skeletonize_frontal_regions(
predicted_region_table=predicted_region_table,
num_grid_rows=num_grid_rows,
num_grid_columns=num_grid_columns)
predicted_label_matrix = object_eval.regions_to_images(
predicted_region_table=predicted_region_table,
num_grid_rows=num_grid_rows,
num_grid_columns=num_grid_columns)
_plot_predictions(predicted_label_matrix=predicted_label_matrix,
title_string='All skeleton lines',
annotation_string='(d)',
output_file_name=ALL_SKELETONS_FILE_NAME)
predicted_region_table = object_eval.find_main_skeletons(
predicted_region_table=predicted_region_table,
image_times_unix_sec=valid_times_unix_sec,
num_grid_rows=num_grid_rows,
num_grid_columns=num_grid_columns,
x_grid_spacing_metres=grid_spacing_metres,
y_grid_spacing_metres=grid_spacing_metres,
min_endpoint_length_metres=MIN_ENDPOINT_LENGTH_METRES)
predicted_label_matrix = object_eval.regions_to_images(
predicted_region_table=predicted_region_table,
num_grid_rows=num_grid_rows,
num_grid_columns=num_grid_columns)
_plot_predictions(predicted_label_matrix=predicted_label_matrix,
title_string='Main skeleton lines',
annotation_string='(e)',
output_file_name=MAIN_SKELETONS_FILE_NAME)
print 'Concatenating figures to: "{0:s}"...'.format(CONCAT_FILE_NAME)
panel_file_names = [
PROBABILITY_FILE_NAME, ALL_REGIONS_FILE_NAME, LARGE_REGIONS_FILE_NAME,
ALL_SKELETONS_FILE_NAME, MAIN_SKELETONS_FILE_NAME
]
imagemagick_utils.concatenate_images(input_file_names=panel_file_names,
output_file_name=CONCAT_FILE_NAME,
num_panel_rows=2,
num_panel_columns=3)
imagemagick_utils.resize_image(input_file_name=CONCAT_FILE_NAME,
output_file_name=CONCAT_FILE_NAME,
output_size_pixels=CONCAT_SIZE_PIXELS)
def _run(input_prediction_dir_name, first_time_string, last_time_string,
num_times, binarization_threshold, min_object_area_metres2,
min_endpoint_length_metres, top_front_line_dir_name,
output_file_name):
"""Converts gridded CNN predictions to objects.
This is effectively the main method.
:param input_prediction_dir_name: See documentation at top of file.
:param first_time_string: Same.
:param last_time_string: Same.
:param num_times: Same.
:param binarization_threshold: Same.
:param min_object_area_metres2: Same.
:param min_endpoint_length_metres: Same.
:param top_front_line_dir_name: Same.
:param output_file_name: Same.
"""
grid_spacing_metres = nwp_model_utils.get_xy_grid_spacing(
model_name=nwp_model_utils.NARR_MODEL_NAME)[0]
num_grid_rows, num_grid_columns = nwp_model_utils.get_grid_dimensions(
model_name=nwp_model_utils.NARR_MODEL_NAME)
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)
possible_times_unix_sec = time_periods.range_and_interval_to_list(
start_time_unix_sec=first_time_unix_sec,
end_time_unix_sec=last_time_unix_sec,
time_interval_sec=NARR_TIME_INTERVAL_SECONDS,
include_endpoint=True)
numpy.random.shuffle(possible_times_unix_sec)
unix_times_sec = []
list_of_predicted_region_tables = []
num_times_done = 0
narr_mask_matrix = None
for i in range(len(possible_times_unix_sec)):
if num_times_done == num_times:
break
this_prediction_file_name = ml_utils.find_gridded_prediction_file(
directory_name=input_prediction_dir_name,
first_target_time_unix_sec=possible_times_unix_sec[i],
last_target_time_unix_sec=possible_times_unix_sec[i],
raise_error_if_missing=False)
if not os.path.isfile(this_prediction_file_name):
continue
num_times_done += 1
unix_times_sec.append(possible_times_unix_sec[i])
print 'Reading data from: "{0:s}"...'.format(this_prediction_file_name)
this_prediction_dict = ml_utils.read_gridded_predictions(
this_prediction_file_name)
class_probability_matrix = this_prediction_dict[
ml_utils.PROBABILITY_MATRIX_KEY]
if narr_mask_matrix is None:
narr_mask_matrix = numpy.invert(
numpy.isnan(class_probability_matrix[0, ..., 0])).astype(int)
# TODO(thunderhoser): This should be a separate method.
class_probability_matrix[..., front_utils.NO_FRONT_INTEGER_ID][
numpy.isnan(class_probability_matrix[
..., front_utils.NO_FRONT_INTEGER_ID])] = 1.
class_probability_matrix[numpy.isnan(class_probability_matrix)] = 0.
print 'Determinizing probabilities...'
this_predicted_label_matrix = object_eval.determinize_probabilities(
class_probability_matrix=this_prediction_dict[
ml_utils.PROBABILITY_MATRIX_KEY],
binarization_threshold=binarization_threshold)
print 'Converting image to frontal regions...'
list_of_predicted_region_tables.append(
object_eval.images_to_regions(
predicted_label_matrix=this_predicted_label_matrix,
image_times_unix_sec=possible_times_unix_sec[[i]]))
print 'Throwing out frontal regions with area < {0:f} km^2...'.format(
METRES2_TO_KM2 * min_object_area_metres2)
list_of_predicted_region_tables[
-1] = object_eval.discard_regions_with_small_area(
predicted_region_table=list_of_predicted_region_tables[-1],
x_grid_spacing_metres=grid_spacing_metres,
y_grid_spacing_metres=grid_spacing_metres,
min_area_metres2=min_object_area_metres2)
print 'Skeletonizing frontal regions...'
list_of_predicted_region_tables[
-1] = object_eval.skeletonize_frontal_regions(
predicted_region_table=list_of_predicted_region_tables[-1],
num_grid_rows=num_grid_rows,
num_grid_columns=num_grid_columns)
list_of_predicted_region_tables[-1] = object_eval.find_main_skeletons(
predicted_region_table=list_of_predicted_region_tables[-1],
image_times_unix_sec=possible_times_unix_sec[[i]],
num_grid_rows=num_grid_rows,
num_grid_columns=num_grid_columns,
x_grid_spacing_metres=grid_spacing_metres,
y_grid_spacing_metres=grid_spacing_metres,
min_endpoint_length_metres=min_endpoint_length_metres)
if num_times_done != num_times:
print '\n'
if len(list_of_predicted_region_tables) == 1:
continue
list_of_predicted_region_tables[-1] = (
list_of_predicted_region_tables[-1].align(
list_of_predicted_region_tables[0], axis=1)[0])
print SEPARATOR_STRING
unix_times_sec = numpy.array(unix_times_sec, dtype=int)
predicted_region_table = pandas.concat(list_of_predicted_region_tables,
axis=0,
ignore_index=True)
predicted_region_table = object_eval.convert_regions_rowcol_to_narr_xy(
predicted_region_table=predicted_region_table,
are_predictions_from_fcn=False)
actual_polyline_table = _read_actual_polylines(
top_input_dir_name=top_front_line_dir_name,
unix_times_sec=unix_times_sec,
narr_mask_matrix=narr_mask_matrix)
print SEPARATOR_STRING
actual_polyline_table = object_eval.project_polylines_latlng_to_narr(
actual_polyline_table)
print 'Writing predicted and observed objects to: "{0:s}"...'.format(
output_file_name)
object_eval.write_predictions_and_obs(
predicted_region_table=predicted_region_table,
actual_polyline_table=actual_polyline_table,
pickle_file_name=output_file_name)