def find_hyperparams(
config_file: str,
train_best_model: bool = typer.Argument(False),
):
search_config = _load_config(config_file, "search")
param_grid = search_config["grid"]
n_jobs = search_config["jobs"]
metric = _load_config(config_file, "metrics")[0]
dummy_hyperparams = {name: {} for name in param_grid.keys()}
estimator = model.build_estimator(dummy_hyperparams)
scoring = metrics.get_scoring_function(metric["name"], **metric["params"])
gs = GridSearchCV(
estimator,
_param_grid_to_sklearn_format(param_grid),
n_jobs=n_jobs,
scoring=scoring,
verbose=3,
)
split = "train"
X, y = _get_dataset(_load_config(config_file, "data"),
splits=[split])[split]
gs.fit(X, y)
hyperparams = _param_grid_to_custom_format(gs.best_params_)
estimator = model.build_estimator(hyperparams)
output_dir = _load_config(config_file, "export")["output_dir"]
_save_versioned_estimator(estimator, hyperparams, output_dir)
def _experiment_fn(run_config, hparams):
# num_epochs can control duration if train_steps isn't
# passed to Experiment
train_input = lambda: model.generate_input_fn(
hparams.train_files,
num_epochs=hparams.num_epochs,
batch_size=hparams.train_batch_size,
)
# Don't shuffle evaluation data
eval_input = lambda: model.generate_input_fn(hparams.eval_files,
batch_size=hparams.
eval_batch_size,
shuffle=False)
return tf.contrib.learn.Experiment(
model.build_estimator(
embedding_size=hparams.embedding_size,
# Construct layers sizes with exponetial decay
hidden_units=[
max(
2,
int(hparams.first_layer_size *
hparams.scale_factor**i))
for i in range(hparams.num_layers)
],
config=run_config),
train_input_fn=train_input,
eval_input_fn=eval_input,
**experiment_args)
def _experiment_fn(run_config, hparams):
# num_epochs can control duration if train_steps isn't
# passed to Experiment
train_input = lambda: model.generate_input_fn(
hparams.train_files,
num_epochs=hparams.num_epochs,
batch_size=hparams.train_batch_size,
)
# Don't shuffle evaluation data
eval_input = lambda: model.generate_input_fn(
hparams.eval_files,
batch_size=hparams.eval_batch_size,
shuffle=False
)
return tf.contrib.learn.Experiment(
model.build_estimator(
embedding_size=hparams.embedding_size,
# Construct layers sizes with exponetial decay
hidden_units=[
max(2, int(hparams.first_layer_size *
hparams.scale_factor**i))
for i in range(hparams.num_layers)
],
config=run_config
),
train_input_fn=train_input,
eval_input_fn=eval_input,
**experiment_args
)
def _experiment_fn(output_dir):
# num_epochs can control duration if train_steps isn't
# passed to Experiment
train_input = model.generate_input_fn(
train_files,
num_epochs=num_epochs,
batch_size=train_batch_size,
)
# Don't shuffle evaluation data
eval_input = model.generate_input_fn(eval_files,
batch_size=eval_batch_size,
shuffle=False)
return tf.contrib.learn.Experiment(
model.build_estimator(
output_dir,
# Construct layers sizes with exponetial decay
hidden_units=[
max(2, int(first_layer_size * scale_factor**i))
for i in range(num_layers)
]),
train_input_fn=train_input,
eval_input_fn=eval_input,
# export strategies control the prediction graph structure
# of exported binaries.
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 train(config_file: str):
hyperparams = _load_config(config_file, "hyperparams")
split = "train"
X, y = _get_dataset(_load_config(config_file, "data"), splits=[split])[split]
estimator = model.build_estimator(hyperparams)
estimator.fit(X, y)
output_dir = _load_config(config_file, "export")["output_dir"]
version = _save_versioned_estimator(estimator, hyperparams, output_dir)
return version
def train_and_maybe_evaluate(hparams):
"""Run the training and evaluate using the high level API.
Args:
hparams: Holds hyperparameters used to train the model as name/value pairs.
Returns:
The estimator that was used for training (and maybe eval)
"""
schema = bookings.read_schema(hparams.schema_file)
tf_transform_output = tft.TFTransformOutput(hparams.tf_transform_dir)
train_input = lambda: model.input_fn(
hparams.train_files,
tf_transform_output,
batch_size=TRAIN_BATCH_SIZE
)
eval_input = lambda: model.input_fn(
hparams.eval_files,
tf_transform_output,
batch_size=EVAL_BATCH_SIZE
)
train_spec = tf.estimator.TrainSpec(
train_input, max_steps=hparams.train_steps)
serving_receiver_fn = lambda: model.example_serving_receiver_fn(
tf_transform_output, schema)
exporter = tf.estimator.FinalExporter('bookings', serving_receiver_fn)
eval_spec = tf.estimator.EvalSpec(
eval_input,
steps=hparams.eval_steps,
exporters=[exporter],
name='bookings-eval')
run_config = tf.estimator.RunConfig(
save_checkpoints_steps=999, keep_checkpoint_max=1)
serving_model_dir = os.path.join(hparams.output_dir, SERVING_MODEL_DIR)
run_config = run_config.replace(model_dir=serving_model_dir)
estimator = model.build_estimator(
tf_transform_output,
# Construct layers sizes with exponetial decay
hidden_units=[
max(2, int(FIRST_DNN_LAYER_SIZE * DNN_DECAY_FACTOR**i))
for i in range(NUM_DNN_LAYERS)
],
config=run_config)
tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)
return estimator
def train_and_maybe_evaluate(train_files, eval_files, hparams):
"""Run the training and evaluate using the high level API.
Args:
hparams: Holds hyperparameters used to train the model as name/value pairs.
Returns:
The estimator that was used for training (and maybe eval)
"""
schema = taxi.read_schema('schema.pbtxt')
train_input = lambda: model.input_fn(
train_files,
hparams.tf_transform_dir,
batch_size=TRAIN_BATCH_SIZE
)
eval_input = lambda: model.input_fn(
eval_files,
hparams.tf_transform_dir,
batch_size=EVAL_BATCH_SIZE
)
train_spec = tf.estimator.TrainSpec(
train_input, max_steps=hparams.train_steps)
serving_receiver_fn = lambda: model.example_serving_receiver_fn(
hparams.tf_transform_dir, schema)
exporter = tf.estimator.FinalExporter('chicago-taxi', serving_receiver_fn)
eval_spec = tf.estimator.EvalSpec(
eval_input,
steps=hparams.eval_steps,
exporters=[exporter],
name='chicago-taxi-eval')
run_config = tf.estimator.RunConfig(
save_checkpoints_steps=999, keep_checkpoint_max=1)
serving_model_dir = os.path.join(hparams.output_dir, SERVING_MODEL_DIR)
run_config = run_config.replace(model_dir=serving_model_dir)
estimator = model.build_estimator(
hparams.tf_transform_dir,
# Construct layers sizes with exponetial decay
hidden_units=[
max(2, int(FIRST_DNN_LAYER_SIZE * DNN_DECAY_FACTOR**i))
for i in range(NUM_DNN_LAYERS)
],
config=run_config)
tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)
return estimator
def _experiment_fn(run_config, hparams):
# num_epochs can control duration if train_steps isn't
# passed to Experiment
train_input = lambda: model.generate_input_fn()
# Don't shuffle evaluation data
eval_input = lambda: model.generate_input_fn(shuffle=False)
return tf.contrib.learn.Experiment(
model.build_estimator(config=run_config),
train_input_fn=train_input,
eval_input_fn=eval_input,
**experiment_args)
def _experiment_fn(output_dir):
return tf.contrib.learn.Experiment(
model.build_estimator(output_dir, n_classes, params, config),
train_input_fn=model.make_input_fn_stroke(
files_pattern=os.path.join(data_dir, 'train-*'),
batch_size=train_batch_size),
eval_input_fn=model.make_input_fn_stroke(
files_pattern=os.path.join(data_dir, 'eval-*'),
batch_size=eval_batch_size),
export_strategies=[
tf.contrib.learn.utils.saved_model_export_utils.
make_export_strategy(model.serving_input_fn, exports_to_keep=1)
],
train_steps=train_steps,
eval_steps=eval_steps,
**experiment_args)
def train_and_evaluate(args):
"""Run the training and evaluate using the high level API"""
train_input = model._make_training_input_fn(
args.tft_working_dir,
args.train_filebase,
num_epochs=args.num_epochs,
batch_size=args.train_batch_size,
buffer_size=args.train_buffer_size,
prefetch_buffer_size=args.train_prefetch_buffer_size)
# Don't shuffle evaluation data.
eval_input = model._make_training_input_fn(
args.tft_working_dir,
args.eval_filebase,
shuffle=False,
batch_size=args.eval_batch_size,
buffer_size=1,
prefetch_buffer_size=args.eval_prefetch_buffer_size)
train_spec = tf.estimator.TrainSpec(
train_input, max_steps=args.train_steps)
exporter = tf.estimator.FinalExporter(
'tft_classifier', model._make_serving_input_fn(args.tft_working_dir))
eval_spec = tf.estimator.EvalSpec(
eval_input,
steps=args.eval_steps,
exporters=[exporter],
name='tft_classifier-eval')
run_config = tf.estimator.RunConfig()
run_config = run_config.replace(model_dir=args.job_dir)
print('model dir {}'.format(run_config.model_dir))
estimator = model.build_estimator(
config=run_config,
tft_working_dir=args.tft_working_dir,
embedding_size=args.embedding_size,
# Construct layers sizes with exponential decay.
hidden_units=[
max(2, int(args.first_layer_size * args.scale_factor**i))
for i in range(args.num_layers)
],
)
tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)
def run_experiment(hparams):
"""Run the training and evaluate using the high level API"""
train_input = model._make_training_input_fn(
hparams.tft_working_dir,
hparams.train_filebase,
num_epochs=hparams.num_epochs,
batch_size=hparams.train_batch_size,
buffer_size=hparams.train_buffer_size,
prefetch_buffer_size=hparams.train_prefetch_buffer_size)
# Don't shuffle evaluation data
eval_input = model._make_training_input_fn(
hparams.tft_working_dir,
hparams.eval_filebase,
shuffle=False,
batch_size=hparams.eval_batch_size,
buffer_size=1,
prefetch_buffer_size=hparams.eval_prefetch_buffer_size)
train_spec = tf.estimator.TrainSpec(train_input,
max_steps=hparams.train_steps)
exporter = tf.estimator.FinalExporter(
'tft_classifier',
model._make_serving_input_fn(hparams.tft_working_dir))
eval_spec = tf.estimator.EvalSpec(eval_input,
steps=hparams.eval_steps,
exporters=[exporter],
name='tft_classifier-eval')
run_config = tf.estimator.RunConfig()
run_config = run_config.replace(model_dir=hparams.job_dir)
print('model dir {}'.format(run_config.model_dir))
estimator = model.build_estimator(
config=run_config,
tft_working_dir=hparams.tft_working_dir,
embedding_size=hparams.embedding_size,
# Construct layers sizes with exponetial decay
hidden_units=[
max(2, int(hparams.first_layer_size * hparams.scale_factor**i))
for i in range(hparams.num_layers)
],
)
tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)
def _experiment_fn(output_dir):
return Experiment(
model.build_estimator(output_dir),
train_input_fn=model.get_input_fn(
filename=os.path.join(data_dir, 'train.tfrecords'),
batch_size=train_batch_size),
eval_input_fn=model.get_input_fn(
filename=os.path.join(data_dir, 'test.tfrecords'),
batch_size=eval_batch_size),
export_strategies=[saved_model_export_utils.make_export_strategy(
model.serving_input_fn,
default_output_alternative_key=None,
exports_to_keep=1)],
train_steps=train_steps,
eval_steps=eval_steps,
**experiment_args
)
def run_experiment(hparams):
"""Run the training and evaluate using the high level API"""
train_input = lambda: model.input_fn(
hparams.train_files,
num_epochs=hparams.num_epochs,
batch_size=hparams.train_batch_size
)
# Don't shuffle evaluation data
eval_input = lambda: model.input_fn(
hparams.eval_files,
batch_size=hparams.eval_batch_size,
shuffle=False
)
train_spec = tf.estimator.TrainSpec(train_input,
max_steps=hparams.train_steps
)
exporter = tf.estimator.FinalExporter('census',
model.SERVING_FUNCTIONS[hparams.export_format])
eval_spec = tf.estimator.EvalSpec(eval_input,
steps=hparams.eval_steps,
exporters=[exporter],
name='census-eval'
)
run_config = tf.estimator.RunConfig()
run_config = run_config.replace(model_dir=hparams.job_dir)
print('model dir {}'.format(run_config.model_dir))
estimator = model.build_estimator(
embedding_size=hparams.embedding_size,
# Construct layers sizes with exponetial decay
hidden_units=[
max(2, int(hparams.first_layer_size *
hparams.scale_factor**i))
for i in range(hparams.num_layers)
],
config=run_config
)
tf.estimator.train_and_evaluate(estimator,
train_spec,
eval_spec)
def _experiment_fn(output_dir):
return tf.contrib.learn.Experiment(
model.build_estimator(output_dir, n_classes, params, config),
train_input_fn=model.make_input_fn(
files_pattern=os.path.join(data_dir, 'train-*'),
batch_size=train_batch_size),
eval_input_fn=model.make_input_fn(
files_pattern=os.path.join(data_dir, 'eval-*'),
batch_size=eval_batch_size),
export_strategies=[
tf.contrib.learn.utils.saved_model_export_utils.make_export_strategy(
model.serving_input_fn,
exports_to_keep=1)
],
train_steps=train_steps,
eval_steps=eval_steps,
**experiment_args
)
def _experiment_fn(output_dir):
train_input = model.generate_input_fn(train_file,
num_epochs=num_epochs,
batch_size=train_batch_size)
eval_input = model.generate_input_fn(eval_file,
batch_size=eval_batch_size)
return Experiment(model.build_estimator(job_dir,
embedding_size=embedding_size,
hidden_units=hidden_units),
train_input_fn=train_input,
eval_input_fn=eval_input,
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(run_config, hparams):
"""Definicion de experimento"""
# Funcion de entrada de entrenamiento
train_input = lambda: model.generate_input_fn(
hparams.train_files,
num_epochs=hparams.num_epochs,
batch_size=hparams.train_batch_size,
)
# Funcion de entrada de evaluacion
eval_input = lambda: model.generate_input_fn(
hparams.eval_files,
batch_size=hparams.eval_batch_size
)
# Experimento
return tf.contrib.learn.Experiment(
model.build_estimator(hparams.job_dir),
train_input_fn=train_input,
eval_input_fn=eval_input,
**experiment_args
)
def _experiment_fn(output_dir):
input_fn = model.generate_csv_input_fn
train_input = input_fn(train_data_paths,
num_epochs=num_epochs,
batch_size=train_batch_size)
eval_input = input_fn(eval_data_paths,
batch_size=eval_batch_size,
mode=tf.contrib.learn.ModeKeys.EVAL)
return Experiment(model.build_estimator(output_dir,
hidden_units=hidden_units),
train_input_fn=train_input,
eval_input_fn=eval_input,
export_strategies=[
saved_model_export_utils.make_export_strategy(
model.serving_input_fn,
default_output_alternative_key=None,
exports_to_keep=1)
],
eval_metrics=model.get_eval_metrics(),
**experiment_args)
def _experiment_fn(output_dir):
input_fn = model.generate_csv_input_fn
train_input = input_fn(
train_data_paths, num_epochs=num_epochs, batch_size=train_batch_size)
eval_input = input_fn(
eval_data_paths, batch_size=eval_batch_size, mode=tf.contrib.learn.ModeKeys.EVAL)
return Experiment(
model.build_estimator(
output_dir,
hidden_units=hidden_units
),
train_input_fn=train_input,
eval_input_fn=eval_input,
export_strategies=[saved_model_export_utils.make_export_strategy(
model.serving_input_fn,
default_output_alternative_key=None,
exports_to_keep=1
)],
eval_metrics=model.get_eval_metrics(),
#min_eval_frequency = 1000, # change this to speed up training on large datasets
**experiment_args
)
def _experiment_fn(output_dir):
input_fn = (model.generate_csv_input_fn if format == 'csv'
else model.generate_tfrecord_input_fn)
train_input = input_fn(
train_data_paths, num_epochs=num_epochs, batch_size=train_batch_size)
eval_input = input_fn(
eval_data_paths, batch_size=eval_batch_size, mode=tf.contrib.learn.ModeKeys.EVAL)
return Experiment(
model.build_estimator(
output_dir,
nbuckets=nbuckets,
hidden_units=parse_to_int(hidden_units)
),
train_input_fn=train_input,
eval_input_fn=eval_input,
export_strategies=[saved_model_export_utils.make_export_strategy(
model.serving_input_fn,
default_output_alternative_key=None,
exports_to_keep=1
)],
eval_metrics=model.get_eval_metrics(),
#min_eval_frequency = 1000, # change this to speed up training on large datasets
**experiment_args
)
def _experiment_fn(output_dir):
train_input = model.generate_input_fn(train_file,
num_epochs=num_epochs,
batch_size=train_batch_size)
eval_input = model.generate_input_fn(eval_file,
batch_size=eval_batch_size)
return learn.Experiment(
model.build_estimator(job_dir,
embedding_size=embedding_size,
hidden_units=hidden_units),
train_input_fn=train_input,
eval_input_fn=eval_input,
eval_metrics={
'training/hptuning/metric':
learn.MetricSpec(metric_fn=metrics.streaming_accuracy,
prediction_key='logits')
},
export_strategies=[
saved_model_export_utils.make_export_strategy(
model.serving_input_fn,
default_output_alternative_key=None,
exports_to_keep=1)
],
**experiment_args)
parser.add_argument(
'--job-dir',
help='GCS location to write checkpoints and export models',
required=True)
# Argument to turn on all logging
parser.add_argument(
'--verbosity',
choices=['DEBUG', 'ERROR', 'FATAL', 'INFO', 'WARN'],
default='INFO',
)
args = parser.parse_args()
run_config = run_config.RunConfig(model_dir=args.job_dir)
estimator = model.build_estimator(run_config)
def prediction_input_fn():
feature_placeholders = {
'wvec': tf.placeholder(tf.float32, [1, 2, 3]),
'dvec': tf.placeholder(tf.float32, [1, 2, 3]),
}
features = {
key: tf.expand_dims(tensor, -1)
for key, tensor in feature_placeholders.items()
}
return tf.contrib.learn.InputFnOps(features, None,
feature_placeholders)
predictor = from_contrib_estimator(estimator=estimator,
# df['signup_date'] = df['signup_date'].apply(lambda x: start + timedelta(days=x))
predict_df['last_service_use_date'] = predict_df[
'last_service_use_date'].apply(lambda x: start + timedelta(days=x))
# df.rename(columns={'Unnamed: 0': 'user_id'}, inplace=True)
# Get user's recency
predict_df['recency'] = predict_df['last_service_use_date'].apply(
lambda x: (predict_df.last_service_use_date.max() - x).days)
# Convert True False to 0 & 1
predict_df.loc[predict_df['business_service'] == True,
'business_service'] = '1'
predict_df.loc[predict_df['business_service'] == False,
'business_service'] = '0'
predict_df['is_retained'] = 0
# df.loc[df['last_service_use_date'].dt.month.isin([6,7]), 'is_retained'] = 1
predict_df.business_service = predict_df.business_service.astype(str)
predict_df.dropna(inplace=True)
m = build_estimator('model_dir')
predicted_values = list(m.predict(input_fn=lambda: input_fn(predict_df)))
probs = list(m.predict_proba(input_fn=lambda: input_fn(predict_df)))
predict_df['predicted_values'] = predicted_values
predict_df['probs'] = probs
predict_df.to_csv('predicttions.csv')
def train_and_evaluate_model(config, hparams):
"""Runs the local training job given provided command line arguments.
Args:
config: RunConfig object
hparams: dictionary passed by command line arguments
"""
with open(hparams['train_file']) as f:
if hparams['trainer_type'] == 'spam':
contents, labels, _ = train_ml_helpers.spam_from_file(f)
else:
contents, labels = train_ml_helpers.component_from_file(f)
logger.info('Training data received. Len: %d' % len(contents))
# Generate features and targets from extracted contents and labels.
if hparams['trainer_type'] == 'spam':
features, targets = train_ml_helpers \
.transform_spam_csv_to_features(contents, labels)
else:
top_list = top_words.make_top_words_list(contents, hparams['job_dir'])
features, targets, index_to_component = train_ml_helpers \
.transform_component_csv_to_features(contents, labels, top_list)
# Split training and testing set.
logger.info('Features generated')
features_train, features_test, targets_train, targets_test = train_test_split(
features, targets, test_size=0.2, random_state=42)
# Generate TrainSpec and EvalSpec for train and evaluate.
estimator = model.build_estimator(config=config,
job_dir=hparams['job_dir'],
trainer_type=hparams['trainer_type'],
class_count=len(set(labels)))
exporter = tf.estimator.LatestExporter(
name='saved_model',
serving_input_receiver_fn=generate_json_input_fn(
hparams['trainer_type']))
train_spec = tf.estimator.TrainSpec(input_fn=make_input_fn(
hparams['trainer_type'],
features_train,
targets_train,
num_epochs=hparams['num_epochs'],
batch_size=hparams['train_batch_size']),
max_steps=hparams['train_steps'])
eval_spec = tf.estimator.EvalSpec(input_fn=make_input_fn(
hparams['trainer_type'],
features_test,
targets_test,
shuffle=False,
batch_size=hparams['eval_batch_size']),
exporters=exporter,
steps=hparams['eval_steps'])
if hparams['trainer_type'] == 'component':
store_component_conversion(hparams['job_dir'], index_to_component)
result = tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)
logging.info(result)