def generate_experiment_fn(output_dir, hparams):
get_train = model.read_dataset(hparams['train_data_paths'],
tf.estimator.ModeKeys.TRAIN,
hparams['train_batch_size'])
get_valid = model.read_dataset(hparams['eval_data_paths'],
tf.estimator.ModeKeys.EVAL,
1000)
estimator = create_custom_estimator(output_dir, hparams)
train_spec = tf.estimator.TrainSpec(input_fn=get_train, max_steps=hparams['train_steps'])
exporter = tf.estimator.LatestExporter('exporter', model.serving_input_fn)
eval_spec = tf.estimator.EvalSpec(input_fn=get_valid,
steps=1000,
start_delay_secs=1, # start evaluating after N seconds
throttle_secs=10, # evaluate every N seconds
exporters=exporter)
return tf.estimator.train_and_evaluate(estimator=estimator, train_spec=train_spec, eval_spec=eval_spec)
def experiment_fn(output_dir):
get_train = model.read_dataset(train_data_paths, mode=tf.contrib.learn.ModeKeys.TRAIN)
get_valid = model.read_dataset(eval_data_paths, mode=tf.contrib.learn.ModeKeys.EVAL)
# run experiment
return tflearn.Experiment(
tflearn.Estimator(model_fn=model.simple_rnn, model_dir=output_dir),
train_input_fn=get_train,
eval_input_fn=get_valid,
eval_metrics={
'rmse': tflearn.MetricSpec(
metric_fn=tf.contrib.metrics.streaming_root_mean_squared_error
)
},
export_strategies=[saved_model_export_utils.make_export_strategy(
model.serving_input_fn,
default_output_alternative_key=None,
exports_to_keep=1
)],
**experiment_args
)
def experiment_fn(output_dir):
get_train = model.read_dataset(hparams['train_data_paths'],
tf.estimator.ModeKeys.TRAIN,
hparams['train_batch_size'])
get_valid = model.read_dataset(hparams['eval_data_paths'],
tf.estimator.ModeKeys.EVAL, 1000)
eval_freq = max(1, min(2000, hparams['train_steps'] / 5))
return tf.contrib.learn.Experiment(
estimator=create_custom_estimator(output_dir, hparams),
train_input_fn=get_train,
eval_input_fn=get_valid,
train_steps=hparams['train_steps'],
eval_steps=1,
min_eval_frequency=eval_freq,
export_strategies=[
saved_model_export_utils.make_export_strategy(
model.serving_input_fn,
default_output_alternative_key=None,
exports_to_keep=1)
])
def experiment_fn(output_dir):
get_train = model.read_dataset(train_data_paths,
mode=tf.contrib.learn.ModeKeys.TRAIN)
get_valid = model.read_dataset(eval_data_paths,
mode=tf.contrib.learn.ModeKeys.EVAL)
# run experiment
return tflearn.Experiment(
tflearn.Estimator(model_fn=model.simple_rnn, model_dir=output_dir),
train_input_fn=get_train,
eval_input_fn=get_valid,
eval_metrics={
'rmse':
tflearn.MetricSpec(metric_fn=tf.contrib.metrics.
streaming_root_mean_squared_error)
},
export_strategies=[
saved_model_export_utils.make_export_strategy(
model.serving_input_fn,
default_output_alternative_key=None,
exports_to_keep=1)
],
**experiment_args)
def experiment_fn(output_dir):
get_train = model.read_dataset(hparams['train_data_paths'],
tf.estimator.ModeKeys.TRAIN,
hparams['train_batch_size'])
get_valid = model.read_dataset(hparams['eval_data_paths'],
tf.estimator.ModeKeys.EVAL,
1000)
eval_freq = max(1, min(2000, hparams['train_steps']/5))
return tf.contrib.learn.Experiment(
estimator=create_custom_estimator(output_dir, hparams),
train_input_fn=get_train,
eval_input_fn=get_valid,
train_steps=hparams['train_steps'],
eval_steps=1,
min_eval_frequency=eval_freq,
export_strategies=[saved_model_export_utils.make_export_strategy(
model.serving_input_fn,
default_output_alternative_key=None,
exports_to_keep=1
)]
)
FIELD_DEFAULTS.append(["NA"])
else:
FIELD_DEFAULTS.append([0.0])
# run
tf.logging.set_verbosity(tf.logging.INFO)
# create estimator
feature_columns, estimator = model.build_model(
COLUMNS, LABEL_FIELD, FIELD_TYPES, FIELD_CATEGORIES, output_dir,
LEARNING_RATE, L1_NORM, L2_NORM, FEATURE_COLUMN_NUM_BUCKETS, MINS,
MAXS)
train_spec = tf.estimator.TrainSpec(input_fn=model.read_dataset(
MODEL_DIR,
FIELD_DEFAULTS,
COLUMNS,
LABEL_FIELD,
BATCH_SIZE=arguments['BATCH_SIZE'],
TRAIN_STEPS=arguments['TRAIN_STEPS'],
mode=tf.estimator.ModeKeys.TRAIN),
max_steps=arguments['TRAIN_STEPS'])
eval_spec = tf.estimator.EvalSpec(
input_fn=model.read_dataset(
MODEL_DIR,
FIELD_DEFAULTS,
COLUMNS,
LABEL_FIELD,
BATCH_SIZE=
300, #arguments['BATCH_SIZE'], # 300 should guarantee the full eval set coverage
TRAIN_STEPS=arguments['TRAIN_STEPS'],
mode=tf.estimator.ModeKeys.EVAL),
elif (FIELD_TYPES[c] == "string"):
FIELD_DEFAULTS.append(["NA"])
else:
FIELD_DEFAULTS.append([0.0])
# run
tf.logging.set_verbosity(tf.logging.INFO)
# create estimator
feature_columns, estimator = model.build_model(
COLUMNS, LABEL_FIELD, FIELD_TYPES, FIELD_CATEGORIES, output_dir,
LEARNING_RATE, L1_NORM, L2_NORM, EMBEDDING_COLUMNS_SIZE, HIDDEN_UNITS)
train_spec = tf.estimator.TrainSpec(input_fn=model.read_dataset(
MODEL_DIR,
FIELD_DEFAULTS,
COLUMNS,
LABEL_FIELD,
BATCH_SIZE=arguments['BATCH_SIZE'],
TRAIN_STEPS=arguments['TRAIN_STEPS'],
mode=tf.estimator.ModeKeys.TRAIN),
max_steps=arguments['TRAIN_STEPS'])
eval_spec = tf.estimator.EvalSpec(
input_fn=model.read_dataset(MODEL_DIR,
FIELD_DEFAULTS,
COLUMNS,
LABEL_FIELD,
BATCH_SIZE=arguments['BATCH_SIZE'],
TRAIN_STEPS=arguments['TRAIN_STEPS'],
mode=tf.estimator.ModeKeys.EVAL),
start_delay_secs=2 * 60, #20 * 60, # start evaluating after N seconds
throttle_secs=1 * 60) #10 * 60) # evaluate every N seconds
{}).get('trial', ''))
# run
tf.logging.set_verbosity(tf.logging.INFO)
# create estimator
estimator = model.wide_and_deep_model(output_dir, arguments['origin_file'],
arguments['dest_file'],
arguments['nbuckets'],
arguments['hidden_units'],
arguments['learning_rate'])
estimator = tf.contrib.estimator.add_metrics(estimator, model.my_rmse)
train_spec = tf.estimator.TrainSpec(input_fn=model.read_dataset(
arguments['traindata'],
mode=tf.estimator.ModeKeys.TRAIN,
batch_size=arguments['batch_size'],
num_training_epochs=arguments['num_training_epochs']))
eval_spec = tf.estimator.EvalSpec(
input_fn=model.read_dataset(arguments['evaldata']),
steps=None,
start_delay_secs=20 * 60, # start evaluating after N seconds
throttle_secs=10 * 60) # evaluate every N seconds
tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)
estimator.export_savedmodel(
os.path.join(output_dir, 'Servo'),
serving_input_receiver_fn=model.serving_input_fn())
