def run_experiment(params):
"""Testbed for running model training and evaluation."""
dataset = inputs.download_data(params.train_path, params.eval_path)
estimator = model.get_estimator(params)
trial_id = _get_trial_id()
model_dir = os.path.join(params.model_dir, trial_id)
_train_and_evaluate(estimator, dataset, model_dir, params)
def run_training(params):
"""Initializes the estimator and runs train_and_evaluate."""
estimator = model.get_estimator(params)
train_input_fn = inputs.get_input_fn(
params.train_path,
shuffle=True,
batch_size=params.batch_size,
num_epochs=params.num_epochs,
)
train_spec = tf.estimator.TrainSpec(
input_fn=train_input_fn,
max_steps=params.max_steps,
)
eval_input_fn = inputs.get_input_fn(
params.eval_path,
shuffle=False,
batch_size=params.batch_size,
)
exporter = tf.estimator.BestExporter(
"export", inputs.get_serving_input_fn(params.export_format),
exports_to_keep=1)
eval_spec = tf.estimator.EvalSpec(
input_fn=eval_input_fn,
throttle_secs=1,
steps=params.eval_steps,
start_delay_secs=1,
exporters=[exporter],
)
tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)
def run_experiment(arguments):
"""Testbed for running model training and evaluation."""
logging.info('Arguments: %s', arguments)
# Get estimator
estimator = model.get_estimator(arguments)
# Run training and evaluation
_train_and_evaluate(estimator, arguments.job_dir)
def run_experiment(flags):
"""Testbed for running model training and evaluation."""
# Get data for training and evaluation
dataset = utils.read_df_from_bigquery(flags.input,
num_samples=flags.num_samples)
# Get model
estimator = model.get_estimator(flags)
# Run training and evaluation
_train_and_evaluate(estimator, dataset, flags.job_dir)
def run_experiment(flags):
"""Testbed for running model training and evaluation."""
# Get data for training and evaluation
dataset = utils.read_df_from_bigquery(
flags.input, num_samples=flags.num_samples)
# Get model
estimator = model.get_estimator(flags)
# Run training and evaluation
_train_and_evaluate(estimator, dataset, flags.job_dir)
def run_experiment(arguments):
"""Testbed for running model training and evaluation."""
# Get data for training and evaluation
logging.info('Arguments: %s', arguments)
dataset = utils.read_df_from_gcs(arguments.input)
# Get estimator
estimator = model.get_estimator(arguments)
# Run training and evaluation
_train_and_evaluate(estimator, dataset, arguments.job_dir)
def run_experiment(hparams):
"""Train and evaluate tf.estimator model"""
print(hparams)
train_spec = tf.estimator.TrainSpec(
input_fn=input_utils._get_train_input_fn(hparams.train_file,
hparams.batch_size,
hparams.num_epochs,
tf.estimator.ModeKeys.TRAIN),
max_steps=hparams.train_steps)
# final_exporter = tf.estimator.FinalExporter('final_exporter', input_utils.serving_input_fn)
eval_spec = tf.estimator.EvalSpec(input_fn=input_utils._get_train_input_fn(
hparams.eval_file, hparams.batch_size, hparams.num_epochs,
tf.estimator.ModeKeys.EVAL))
# exporters=[final_exporter])
# Checkpoints to save
run_config = tf.estimator.RunConfig(model_dir=hparams.job_dir,
save_checkpoints_steps=100,
keep_checkpoint_max=200)
estimator = model.get_estimator(hparams)
return tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)