| from __future__ import print_function | |
| from __future__ import division | |
| from keras.models import load_model | |
| import argparse | |
| #from sklearn.metrics import f1_score | |
| from datetime import datetime | |
| from tensorflow.python.lib.io import file_io | |
| import h5py | |
| import joblib | |
| """to run code locally: | |
| python test_model.py --job-dir ./ --train-file test_random_shapes.pkl | |
| """ | |
| def train_model(train_file = 'test_resized_images.pkl', | |
| job_dir = './', | |
| **args): | |
| # set the loggining path for ML Engine logging to storage bucket | |
| logs_path = job_dir + '/logs/' + datetime.now().isoformat() | |
| print('Using logs_path located at {}'.format(logs_path)) | |
| # need tensorflow to open file descriptor for google cloud to read | |
| with file_io.FileIO(train_file, mode='r') as f: | |
| # joblib loads compressed files consistenting of large datasets | |
| save = joblib.load(f) | |
| test_shape_dataset = save['train_shape_dataset'] | |
| test_y_dataset = save['train_y_dataset'] | |
| del save # help gc free up memory | |
| # this makes predictions of the model | |
| # the model contains the model architecture and weights, specification of the chosen loss | |
| # and optimization algorithm so that you can resume training if needed | |
| model = load_model('model_ver2.h5') | |
| '''predictions = model.predict(test_shape_dataset, batch_size = 32) | |
| predictions[predictions >= 0.6] = 1 | |
| predictions[predictions < 0.6] = 0 | |
| print ("Label predictions", predictions) | |
| predict_score = f1_score(test_y_dataset, predictions, average='macro') | |
| print("Prediction score", predict_score)''' | |
| # evaluate the model | |
| score = model.evaluate(test_shape_dataset, | |
| test_y_dataset, | |
| batch_size = 32, | |
| verbose = 1) | |
| print ("Test loss:", score[0]) | |
| print ("Test accuracy", score[1]) | |
| print ("Model Summary", model.summary()) | |
| if __name__ == '__main__': | |
| # Parse the input arguments for common Cloud ML Engine options | |
| parser = argparse.ArgumentParser() | |
| parser.add_argument('--train-file', | |
| help='local path of pickle file') | |
| parser.add_argument('--job-dir', | |
| help='Cloud storage bucket to export the model') | |
| args = parser.parse_args() | |
| arguments = args.__dict__ | |
| train_model(**arguments) | |