def run(dataset_name,
model_dir,
batch_size,
predictions_per_example,
max_examples,
output_dir,
fake_data=False):
"""Runs predictions on the given dataset using the specified model."""
gfile.makedirs(output_dir)
data_config = image_data_utils.get_data_config(dataset_name)
dataset = data_lib.build_dataset(data_config,
batch_size,
fake_data=fake_data)
if max_examples:
dataset = dataset.take(max_examples)
model_opts = experiment_utils.load_config(model_dir +
'/model_options.json')
model_opts = models_lib.ModelOptions(**model_opts)
logging.info('Loaded model options: %s', model_opts)
model = models_lib.build_model(model_opts)
logging.info('Loading model weights...')
model.load_weights(model_dir + '/model.ckpt')
logging.info('done loading model weights.')
writer = array_utils.StatsWriter(
os.path.join(output_dir, 'predictions_%s' % dataset_name))
writer_small = array_utils.StatsWriter(
os.path.join(output_dir, 'predictions_small_%s' % dataset_name))
writers = {'full': writer, 'small': writer_small}
max_batches = 50000 // batch_size
experiment_utils.make_predictions(model, dataset, predictions_per_example,
writers, max_batches)
def run(dataset_name,
model_dir,
predictions_per_example,
max_examples,
output_dir,
fake_data=False):
"""Runs predictions on the given dataset using the specified model."""
tf.io.gfile.makedirs(output_dir)
data_config = image_data_utils.get_data_config(dataset_name)
dataset = data_lib.build_dataset(data_config, fake_data=fake_data)
if max_examples:
dataset = dataset.take(max_examples)
model = models_lib.load_model(model_dir)
logging.info('Starting predictions.')
predictions = experiment_utils.make_predictions(model,
dataset.batch(_BATCH_SIZE),
predictions_per_example)
logging.info('Done computing predictions; recording results to disk.')
array_utils.write_npz(output_dir, 'predictions_%s.npz' % dataset_name,
predictions)
del predictions['logits_samples']
array_utils.write_npz(output_dir,
'predictions_small_%s.npz' % dataset_name,
predictions)
