# compile model
model.compile(loss='mse', optimizer='sgd')
example_input = np.array([[1, 2, 3, 4, 5]])
if args.train:
example_target = np.array([[1, 2]])
model.fit(example_input, example_target, epochs=5)
keras_predictions = model.predict(example_input)[0]
# save the weights
model.save(weights_file)
# convert h5 file to txt
h5_to_txt(weights_file_name=weights_file, output_file_name=None)
txt_to_h5(weights_file_name=weights_file.replace('.h5', '.txt'),
output_file_name=None)
cmd = ['../../build/bin/./test_keras', weights_file.replace('.h5', '.txt')]
print('Keras predictions: ', keras_predictions)
if os.path.exists(weights_file.replace('.h5', '_converted.h5')):
model2 = load_model(weights_file.replace('.h5', '_converted.h5'))
keras_predictions2 = model2.predict(example_input)[0]
print('Other Keras predictions:', keras_predictions2)
# run
result = subprocess.check_output(cmd, stderr=subprocess.STDOUT)
fortran_predictions = np.array([float(num) for num in result.strip().split()],
model_path = args.model_dir + file_name
output_path = args.model_dir + file_name.replace('.h5', '.txt')
print('Model path:', model_path)
print('Output path:', output_path)
# load keras model from h5
try:
model = load_model(model_path)
except:
print('Failed...')
continue
keras_predictions = model.predict(sample_input)[0]
# convert h5 file to txt
h5_to_txt(weights_file_name=model_path, output_file_name=output_path)
cmd = ['../../build/bin/./test_bulk', output_path]
# print('Keras predictions:', keras_predictions)
# run
result = subprocess.check_output(cmd, stderr=subprocess.STDOUT)
fortran_predictions = np.array(
[float(num) for num in result.strip().split()], dtype=np.float32)
# print('Fortran predictions:', fortran_predictions)
# keras predictions must match neural fortran predictions
assert np.allclose(keras_predictions, fortran_predictions, atol=1e-3)
output2 = Dense(1)(x)
output3 = Dense(1)(x)
# CREATE THE MODEL
multi_output_model = Model(input,outputs=[output1, output2, output3])
multi_output_model.compile(
loss='categorical_crossentropy',
optimizer='sgd',
metrics=['accuracy']
)
# SAVE TO FILE FOR PARSING
multi_output_model.save('multi_output_model.h5')
# CONVERT TO TXT
convert_weights.h5_to_txt('multi_output_model.h5', 'single_output_model.txt')
# CONVERT TO H5
convert_weights.txt_to_h5('single_output_model.txt', 'single_output_model.h5')
single_output_model = load_model('single_output_model.h5')
# GRAPHIC PLOT OF MODEL
plot_model(multi_output_model, to_file='../../Figures/multi_output_model.png', show_shapes=True, show_layer_names=True)
plot_model(single_output_model, to_file='../../Figures/single_output_model.png', show_shapes=True, show_layer_names=True)
# TEST INPUT
input = np.array(
[[1,2,3,4,5]]
)
# COMPARE PREDICTIONS FROM MULTI OUTPUT AND SINGLE OUTPUT MODELS
# plt.plot(t['pred'].tolist(), label='prediction')
# plt.ylabel('output')
# plt.legend()
# plt.tight_layout()
# plt.savefig("adim_regression.pdf", dpi=150)
# plt.show()
# plt.close()
#
## Plot the chart
#chart_regression(pred.flatten(), y_test)
#model.save('model.sav')
dump(model, 'model.sav')
#
model_name_h5 = 'NN.h5'
model_name_txt = model_name_h5.replace('h5', 'txt')
model.save(model_name_h5, include_optimizer=False)
h5_to_txt(model_name_h5, model_name_txt)
#plot_model(model, to_file="./model.png", show_shapes=True, show_layer_names=True)
#tf.keras.utils.plot_model(model, to_file='model.png', show_shapes=True, show_layer_names=True)
#new_model = tf.keras.models.load_model('model.sav')
#new_model.summary()
#keras_file = 'linear.h5'
#keras.models.save_model(model, keras_file)
#
#converter = lite.TocoConverter.from_keras_model_file(keras_file)
#tflite_model = converter.convert()
#open('linear.tflite', 'wb').write(tflite_model)
y_test = keras.utils.to_categorical(y_test, num_classes)
# SEQUENTIAL MODEL
model = Sequential()
model.add(Dense(512, activation='relu', input_shape=(784, )))
model.add(Dense(512, activation='relu'))
model.add(Dense(num_classes, activation='softmax'))
# FUNCTIONAL MODEL
# input = x = Input(shape=(784,))
# x = Dense(512, activation='relu')(x)
# x = Dense(512, activation='relu')(x)
# x = Dense(num_classes, activation='softmax')(x)
#
# model = Model(inputs=input, outputs=x)
model.compile(loss='categorical_crossentropy',
optimizer=RMSprop(),
metrics=['accuracy'])
history = model.fit(x_train,
y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(x_test, y_test))
model.save(weights_file_name)
h5_to_txt(weights_file_name, txt_file_name)
