def print_desc(self, resume_from):
desc = f"NAME: {self.name}\n"
desc += f"DATA_DIR: {self.data_dir}\n"
desc += f"BATCH_SIZE: {self.BATCH_SIZE}\n"
desc += f"BUFFER_SIZE: {self.BUFFER_SIZE}\n"
desc += f"PREFETCH_SIZE: {self.PREFETCH_SIZE}\n"
desc += f"NB_BATCHES: {self.NB_BATCHES}\n"
desc += f"SAMPLE_SHAPE: {self.sample_shape}\n"
desc += f"OUTPUT_SHAPE: {self.output_shape}\n"
desc += f"MIN_WIDTH/HEIGHT: {self.min_wh}\n"
desc += f"MIN_WEIGHT: {self.min_wei} == {2 ** self.min_wei}\n"
desc += f"NB_LAYERS: {self.nb_layers}\n"
desc += f"LEARNING_RATE_D: {self.LEARNING_RATE_D}\n"
desc += f"LEARNING_RATE_G: {self.LEARNING_RATE_G}\n"
desc += f"BETA_1: {self.BETA_1}\n"
desc += f"BETA_2: {self.BETA_2}\n"
desc += f"TRAINING_RATIO: {self.TRAINING_RATIO}\n"
desc += f"GRADIENT_PENALTY_WEIGHT: {self.GRADIENT_PENALTY_WEIGHT}\n"
desc += f"Z_SIZE: {self.Z_SIZE}\n"
desc += f"LR_ALPHA: {self.LR_ALPHA}\n"
desc += f"BN_MOMENTUM: {self.BN_MOMENTUM}\n"
desc += f"KERNEL_SIZE: {self.KERNEL_SIZE}\n"
print(desc)
with self.writer.as_default():
tf.summary.trace_on()
text("Hyper-parameters", desc, step=resume_from)
self.writer.flush()
def log_test_results(cfg, model, test_generator, test_metrics, log_dir):
'''
Visualize performance of a trained model on the test set. Optionally save the model.
:param cfg: Project config
:param model: A trained Keras model
:param test_generator: A Keras generator for the test set
:param test_metrics: Dict of test set performance metrics
:param log_dir: Path to write TensorBoard logs
'''
# Visualization of test results
test_predictions = model.predict(test_generator, verbose=0)
test_labels = test_generator.labels
plt = plot_roc(test_labels,
test_predictions,
list(test_generator.class_indices.keys()),
dir_path=cfg['PATHS']['IMAGES'])
roc_img = plot_to_tensor()
plt = plot_confusion_matrix(test_labels,
test_predictions,
list(test_generator.class_indices.keys()),
dir_path=cfg['PATHS']['IMAGES'])
cm_img = plot_to_tensor()
# Log test set results and plots in TensorBoard
writer = tf_summary.create_file_writer(logdir=log_dir)
# Create table of test set metrics
test_summary_str = [['**Metric**', '**Value**']]
for metric in test_metrics:
metric_values = test_metrics[metric]
test_summary_str.append([metric, str(metric_values)])
# Create table of model and train hyperparameters used in this experiment
hparam_summary_str = [['**Variable**', '**Value**']]
for key in cfg['TRAIN']:
hparam_summary_str.append([key, str(cfg['TRAIN'][key])])
for key in cfg['NN'][cfg['TRAIN']['MODEL_DEF'].upper()]:
hparam_summary_str.append(
[key, str(cfg['NN'][cfg['TRAIN']['MODEL_DEF'].upper()][key])])
# Write to TensorBoard logs
with writer.as_default():
tf_summary.text(name='Test set metrics',
data=tf.convert_to_tensor(test_summary_str),
step=0)
tf_summary.text(name='Run hyperparameters',
data=tf.convert_to_tensor(hparam_summary_str),
step=0)
tf_summary.image(name='ROC Curve (Test Set)', data=roc_img, step=0)
tf_summary.image(name='Confusion Matrix (Test Set)',
data=cm_img,
step=0)
return
def log_test_results(cfg, model, test_generator, test_metrics, log_dir):
'''
Visualize performance of a trained model on the test set. Optionally save the model.
:param cfg: Project config
:param model: A trained Keras model
:param test_generator: A Keras generator for the test set
:param test_metrics: Dict of test set performance metrics
:param log_dir: Path to write TensorBoard logs
'''
# Visualization of test results
test_predictions = model.predict_generator(test_generator, verbose=0)
test_labels = test_generator.labels
covid_idx = test_generator.class_indices['COVID-19']
plt = plot_roc("Test set", test_labels, test_predictions, class_id=covid_idx)
roc_img = plot_to_tensor()
plt = plot_confusion_matrix(test_labels, test_predictions, class_id=covid_idx)
cm_img = plot_to_tensor()
# Log test set results and plots in TensorBoard
writer = tf_summary.create_file_writer(logdir=log_dir)
# Create table of test set metrics
test_summary_str = [['**Metric**','**Value**']]
thresholds = cfg['TRAIN']['THRESHOLDS'] # Load classification thresholds
for metric in test_metrics:
if metric in ['precision', 'recall'] and isinstance(metric, list):
metric_values = dict(zip(thresholds, test_metrics[metric]))
else:
metric_values = test_metrics[metric]
test_summary_str.append([metric, str(metric_values)])
# Create table of model and train config values
hparam_summary_str = [['**Variable**', '**Value**']]
for key in cfg['TRAIN']:
hparam_summary_str.append([key, str(cfg['TRAIN'][key])])
if cfg['TRAIN']['CLASS_MODE'] == 'binary':
for key in cfg['NN']['DCNN_BINARY']:
hparam_summary_str.append([key, str(cfg['NN']['DCNN_BINARY'][key])])
else:
for key in cfg['NN']['DCNN_BINARY']:
hparam_summary_str.append([key, str(cfg['NN']['DCNN_BINARY'][key])])
# Write to TensorBoard logs
with writer.as_default():
tf_summary.text(name='Test set metrics', data=tf.convert_to_tensor(test_summary_str), step=0)
tf_summary.text(name='Run hyperparameters', data=tf.convert_to_tensor(hparam_summary_str), step=0)
tf_summary.image(name='ROC Curve (Test Set)', data=roc_img, step=0)
tf_summary.image(name='Confusion Matrix (Test Set)', data=cm_img, step=0)
return