def run(state,
include=None,
exclude=None,
no_reverse=False,
raw=False,
place=None):
"""
Run transformations on data loaded in MongoDB.
State is required. Place is optional.
Optionally provide to limit transforms that are performed.
"""
try:
run_transforms = _select_transforms(state, include, exclude, raw,
place)
except IncludeExcludeError as e:
sys.exit(e)
for transform in run_transforms:
if not no_reverse and transform.auto_reverse:
# Reverse the transform if it's been run previously
transform.reverse()
print 'Executing %s' % transform
transform()
validators = transform.validators.values()
if validators:
print "Executing validation"
run_validation(state, validators)
def run(state, include=None, exclude=None, no_reverse=False):
"""
Run transformations on data loaded in MongoDB.
State is required. Optionally provide to limit transforms that are performed.
"""
try:
run_transforms = _select_transforms(state, include, exclude)
except IncludeExcludeError as e:
sys.exit(e)
for transform in run_transforms:
if not no_reverse and transform.auto_reverse:
# Reverse the transform if it's been run previously
transform.reverse()
print 'Executing %s' % transform
transform()
validators = transform.validators.values()
if validators:
print "Executing validation"
run_validation(state, validators)
model.add(
keras.layers.Dense(32,
activation='relu',
kernel_regularizer=keras.regularizers.l2(0.0001)))
model.add(keras.layers.Dropout(0.4))
model.add(
keras.layers.Dense(
1,
activation='sigmoid',
bias_initializer=keras.initializers.Constant(output_bias)))
model.compile(loss=keras.losses.BinaryCrossentropy(),
optimizer="adam",
metrics=metrics)
print(model.summary())
return model
model = build_model(
input_shape=(40, 40, 3),
filters=filters,
batch_norm=batch_norm,
#output_bias=output_bias,
metrics=metrics)
model.load_weights(weightsDir + 'lastEpoch.h5')
validate.run_validation(model, weightsDir + 'lastEpoch.h5', outputDir, dataDir,
plotDir)
keras.callbacks.EarlyStopping(patience=patience_count),
keras.callbacks.ModelCheckpoint(filepath=weightsDir+'model.{epoch}.h5',
save_best_only=True,
monitor=monitor,
mode='auto')
# tf.keras.callbacks.TensorBoard(log_dir=logDir,
# histogram_freq=0,
# write_graph=False,
# write_images=False)
]
history = model.fit(train_generator,
epochs = epochs,
verbose= v,
validation_data=val_generator,
callbacks=callbacks)
model.save_weights(weightsDir+'lastEpoch.h5')
print(utils.bcolors.GREEN+"Saved weights to "+weightsDir+utils.bcolors.ENDC)
# save and plot history file
with open(outputDir+'history.pkl', 'wb') as f:
pickle.dump(history.history, f)
print(utils.bcolors.GREEN+"Saved history, train and validation files to "+outputDir+utils.bcolors.ENDC)
utils.plot_history(history, plotDir, ['loss','accuracy'])
print(utils.bcolors.YELLOW+"Plotted history to "+plotDir+utils.bcolors.ENDC)
if(run_validate): validate.run_validation(model, weightsDir+'lastEpoch.h5', outputDir, dataDir, plotDir, batch_size)
def validate():
run_validation()
def validate():
run_validation()