def run():
tf.reset_default_graph()
train_dataset, val_dataset = load_all_datasets()
handle, training_iterator, validation_iterator, next_element = load_iterators(train_dataset, val_dataset)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
training_handle = sess.run(training_iterator.string_handle())
validation_handle = sess.run(validation_iterator.string_handle())
i = 0
while True:
try:
sess.run(next_element[0], feed_dict={handle: training_handle})
i+=BATCH_SIZE
except:
print("Train examples: {} +/- {}".format(i, BATCH_SIZE))
break
i = 0
while True:
try:
sess.run(next_element[0], feed_dict={handle: validation_handle})
i+=BATCH_SIZE
except:
print("Validation examples: {} +/- {}".format(i, BATCH_SIZE))
break
def run():
tf.reset_default_graph()
train_dataset, val_dataset = load_all_datasets()
handle, training_iterator, validation_iterator, next_element = load_iterators(
train_dataset, val_dataset)
training, img, labels, out, merged, upd = model(*next_element)
saver = tf.train.Saver(max_to_keep=2)
sess = tf.Session()
train_writer = tf.summary.FileWriter('./logs/train', sess.graph)
val_writer = tf.summary.FileWriter('./logs/val', sess.graph)
sess.run(tf.global_variables_initializer())
training_handle = sess.run(training_iterator.string_handle())
validation_handle = sess.run(validation_iterator.string_handle())
i = 0
spinner = Halo(text='Training', spinner='dots')
subprocess.Popen([
"tensorboard", "--logdir", "./logs", "--port", "6006", "--host",
"0.0.0.0"
])
spinner.start()
# img_out, labels_out = sess.run([img, labels], feed_dict={handle: training_handle})
# img_out = img_out[0].squeeze()
# labels_out = labels_out[0].squeeze()
# import nibabel as nib
# nib.save(nib.Nifti1Image(img_out, np.eye(4)), "original.nii")
# nib.save(nib.Nifti1Image(labels_out, np.eye(4)), "seg.nii")
while True:
m, _ = sess.run([merged, upd], feed_dict={handle: training_handle})
train_writer.add_summary(m, i)
if i % 100 == 0:
m = sess.run(merged,
feed_dict={
training: False,
handle: validation_handle
})
val_writer.add_summary(m, i)
val_writer.flush()
if i % 1000 == 0 and i > 0:
# Save model
saver.save(sess, "./models/model.ckpt", global_step=i)
i += 1
spinner.stop()
def main(experiment_config):
"""Sets the configurations according to `experiment_config` and runs them.
"""
raw_output_filepath = os.path.join(experiment_config["output folder"],
'raw_output.csv')
with open(raw_output_filepath, 'w') as fout:
init_raw_output_csv(fout, output_split_char=',')
criteria_list = get_criteria(experiment_config["criteria"])
if "starting seed index" not in experiment_config:
starting_seed = 1
else:
starting_seed = experiment_config["starting seed index"]
if experiment_config["prunning parameters"]["use chi-sq test"]:
max_p_value_chi_sq = experiment_config["prunning parameters"][
"max chi-sq p-value"]
decision_tree.MIN_SAMPLES_IN_SECOND_MOST_FREQUENT_VALUE = experiment_config[
"prunning parameters"]["second most freq value min samples"]
else:
max_p_value_chi_sq = None
decision_tree.MIN_SAMPLES_IN_SECOND_MOST_FREQUENT_VALUE = None
decision_tree.USE_MIN_SAMPLES_SECOND_LARGEST_CLASS = experiment_config[
"prunning parameters"]["use second most freq class min samples"]
if decision_tree.USE_MIN_SAMPLES_SECOND_LARGEST_CLASS:
decision_tree.MIN_SAMPLES_SECOND_LARGEST_CLASS = experiment_config[
"prunning parameters"]["second most freq class min samples"]
else:
decision_tree.MIN_SAMPLES_SECOND_LARGEST_CLASS = None
if experiment_config["use all datasets"]:
datasets_configs = dataset.load_all_configs(
experiment_config["datasets basepath"])
datasets_configs.sort(key=lambda config: config["dataset name"])
else:
datasets_folders = [
os.path.join(experiment_config["datasets basepath"],
folderpath)
for folderpath in experiment_config["datasets folders"]
]
datasets_configs = [
dataset.load_config(folderpath)
for folderpath in datasets_folders
]
if experiment_config["load one dataset at a time"]:
datasets = dataset.load_all_datasets(
datasets_configs, experiment_config["use numeric attributes"])
for ((dataset_name, curr_dataset),
min_num_samples_allowed) in itertools.product(
datasets, experiment_config["prunning parameters"]
["min num samples allowed"]):
for criterion in criteria_list:
print('-' * 100)
print(criterion.name)
print()
run(dataset_name,
curr_dataset,
experiment_config["num training samples"],
criterion,
min_num_samples_allowed=min_num_samples_allowed,
max_depth=experiment_config["max depth"],
num_trials=experiment_config["num trials"],
starting_seed=starting_seed,
use_numeric_attributes=experiment_config[
"use numeric attributes"],
use_chi_sq_test=experiment_config[
"prunning parameters"]["use chi-sq test"],
max_p_value_chi_sq=max_p_value_chi_sq,
output_file_descriptor=fout,
output_split_char=',')
else:
for (dataset_config, min_num_samples_allowed) in itertools.product(
datasets_configs, experiment_config["prunning parameters"]
["min num samples allowed"]):
curr_dataset = dataset.Dataset(
dataset_config["filepath"],
dataset_config["key attrib index"],
dataset_config["class attrib index"],
dataset_config["split char"],
dataset_config["missing value string"],
experiment_config["use numeric attributes"])
for criterion in criteria_list:
print('-' * 100)
print(criterion.name)
print()
run(dataset_config["dataset name"],
curr_dataset,
experiment_config["num training samples"],
criterion,
min_num_samples_allowed=min_num_samples_allowed,
max_depth=experiment_config["max depth"],
num_trials=experiment_config["num trials"],
starting_seed=starting_seed,
use_numeric_attributes=experiment_config[
"use numeric attributes"],
use_chi_sq_test=experiment_config[
"prunning parameters"]["use chi-sq test"],
max_p_value_chi_sq=max_p_value_chi_sq,
output_file_descriptor=fout,
output_split_char=',')