workers=ngpu,
callbacks=callback_list,
verbose=1)
################ Test with best trained models
pmodel.load_weights(output_best_model)
# Compile model (required to make predictions)
pmodel.compile(loss='binary_crossentropy',
optimizer='rmsprop',
metrics=['accuracy', ef.fbeta_score])
flog.write('best model monitored')
ef.evaluate(pmodel,
test_generator1,
flog=flog,
name=file_test_label1,
output_file=file_test_label1.replace('.mat', '_sepcnn_output.txt'))
ef.evaluate(pmodel,
test_generator2,
flog=flog,
name=file_test_label2,
output_file=file_test_label2.replace('.mat', '_sepcnn_output.txt'))
ef.evaluate(pmodel,
test_generator3,
flog=flog,
name=file_test_label3,
output_file=file_test_label3.replace('.mat', '_sepcnn_output.txt'))
pmodel.save(output_model_cnn)
pmodel.save_weights(output_best_model)
################## Testing
parallel_model.load_weights(output_best_model)
parallel_model.compile(loss='binary_crossentropy', optimizer='rmsprop',
metrics=['accuracy', ef.fbeta_score])
print('best model monitored')
flog.write('best model monitored')
parallel_model.save(output_model_cnn)
parallel_model.save_weights(output_best_model)
model.save(output_model_cnn.replace('.h5', '_1gpu.h5'))
model.save_weights(output_best_model.replace('.h5', '_1gpu.h5'))
ef.evaluate(parallel_model, test_generator)
#########
flog.close()
callback_list = [checkpoint, roc, early_stopping]
trained_history = pmodel.fit_generator(generator=train_generator,
epochs=epoch,
shuffle=True,
validation_data=val_generator,
use_multiprocessing=False,
workers=1,
callbacks=callback_list,
verbose=1)
############## Testing
pmodel.load_weights(output_best_model)
pmodel.compile(loss='binary_crossentropy',
optimizer='rmsprop',
metrics=['accuracy', ef.fbeta_score])
print('best model monitored')
flog.write('best model monitored')
ef.evaluate(pmodel, test_generator, flog=flog)
pmodel.save(output_model)
pmodel.save_weights(output_best_model)
model.save(output_model.replace('.h5', '_1gpu.h5'))
model.save_weights(output_best_model.replace('.h5', '_1gpu.h5'))
#########
flog.close()