def main(argv):
parser = argparse.ArgumentParser(description='TODO')
parser.add_argument('config',
type=str,
help='Path to configuration file to use.')
parser.add_argument(
'mean',
type=str,
help='Path to npy file holding mean for normalization.')
parser.add_argument(
'variance',
type=str,
help='Path to npy file holding variance for normalization.')
parser.add_argument('model_dir',
type=str,
help='Path to saved model to use for inference.')
args = parser.parse_args()
mean = np.load(args.mean)
variance = np.load(args.variance)
stddev = [np.math.sqrt(x) for x in variance]
def _normalize_op(features):
channels = [
tf.expand_dims((features['patch'][:, :, channel] - mean[channel]) /
stddev[channel], -1) for channel in range(3)
]
features['patch'] = tf.concat(channels, 2)
return features
cutil.make_directory(args.model_dir)
cutil.publish(args.model_dir)
config_path = args.config
config = ctfm.parse_json(config_path)
config_datasets = config.get('datasets')
config_model = config.get('model')
train_fn = ctfd.construct_train_fn(config_datasets,
operations=[_normalize_op])
steps = int(
config_datasets.get('training').get('size') /
config_datasets.get('batch'))
params_dict = {
'config': config_model,
'model_dir': args.model_dir,
'mean': mean,
'stddev': stddev
}
classifier = tf.estimator.Estimator(model_fn=my_model,
model_dir=args.model_dir,
params=params_dict,
config=tf.estimator.RunConfig(
model_dir=args.model_dir,
save_summary_steps=1000,
log_step_count_steps=1000))
if not os.path.exists(
os.path.join(args.model_dir, os.path.basename(config_path))):
shutil.copy2(config_path, args.model_dir)
for epoch in range(config_datasets.get('training').get('epochs')):
classifier = classifier.train(input_fn=train_fn, steps=steps)
export_dir = os.path.join(args.model_dir, 'saved_model')
cutil.make_directory(export_dir)
cutil.publish(export_dir)
cutil.publish(args.model_dir)
cutil.publish(export_dir)
def main(argv):
parser = argparse.ArgumentParser(description='TODO')
parser.add_argument('config',
type=str,
help='Path to configuration file to use.')
parser.add_argument(
'mean',
type=str,
help='Path to npy file holding mean for normalization.')
parser.add_argument(
'variance',
type=str,
help='Path to npy file holding variance for normalization.')
parser.add_argument('model_dir',
type=str,
help='Path to saved model to use for inference.')
args = parser.parse_args()
mean = np.load(args.mean)
variance = np.load(args.variance)
stddev = [np.math.sqrt(x) for x in variance]
def _normalize_op(features):
channels = [
tf.expand_dims((features['patch'][:, :, channel] - mean[channel]) /
stddev[channel], -1) for channel in range(3)
]
features['patch'] = tf.concat(channels, 2)
return features
def _subsampling_op(features):
features['patch'] = ctfi.subsample(features['patch'], 2)
return features
cutil.make_directory(args.model_dir)
cutil.publish(args.model_dir)
config_path = args.config
config = ctfm.parse_json(config_path)
config_datasets = config.get('datasets')
config_model = config.get('model')
train_fn = ctfd.construct_train_fn(config_datasets,
operations=[_normalize_op])
#def train_fn():
# dataset = tf.data.Dataset.from_tensor_slices(np.random.rand(256,32,32,3))
# dataset = dataset.map(lambda x : ({"patch": x}, 0)).batch(256).repeat()
# return dataset
steps = int(
config_datasets.get('training').get('size') /
config_datasets.get('batch'))
params_dict = {'config': config_model, 'model_dir': args.model_dir}
classifier = tf.estimator.Estimator(model_fn=my_model,
model_dir=args.model_dir,
params=params_dict,
config=tf.estimator.RunConfig(
model_dir=args.model_dir,
save_summary_steps=1000,
log_step_count_steps=1000))
if not os.path.exists(
os.path.join(args.model_dir, os.path.basename(config_path))):
shutil.copy2(config_path, args.model_dir)
for epoch in range(config_datasets.get('training').get('epochs')):
classifier = classifier.train(input_fn=train_fn, steps=steps)
export_dir = os.path.join(args.model_dir, 'saved_model')
cutil.make_directory(export_dir)
cutil.publish(export_dir)
# TODO: Write command to create serving input receiver fn from config.
serving_input_receiver_fn = ctfd.construct_serving_fn(
config_model['inputs'])
classifier.export_saved_model(export_dir, serving_input_receiver_fn)
cutil.publish(args.model_dir)
cutil.publish(export_dir)