def main(unused_argv):
params = {}
params["PIPELINE_FILE"] = 'pipeline.config'
params["MODEL_DIR"] = 'model.ckpt' #output directory
params["NUM_STEPS"] = int(os.getenv("CK_NUM_STEPS", '1'))
params["EVAL_TRAIN_DATA"] = os.getenv("CK_EVAL_TRAIN_DATA", False)
params["SAMPLE_1_OF_N_EVAL_EXAMPLES"] = int(
os.getenv("CK_SAMPLE_1_OF_N_EVAL_EXAMPLES", 1))
params["SAMPLE_1_OF_N_TRAIN_EXAMPLES"] = int(
os.getenv("CK_SAMPLE_1_OF_N_TRAIN_EXAMPLES", 5))
params["HYPERPARAMS_OVERRIDE"] = os.getenv("CK_HYPERPARAMS_OVERRIDE", None)
params["CHECKPOINT_DIR"] = os.getenv("CK_CHECKPOINT_DIR", None)
params["RUN_ONCE"] = os.getenv("CK_RUN_ONCE", None)
#flags.mark_flag_as_required('model_dir')
#flags.mark_flag_as_required('pipeline_config_path')
config = tf.estimator.RunConfig(params["MODEL_DIR"])
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(params["HYPERPARAMS_OVERRIDE"]),
pipeline_config_path=params["PIPELINE_FILE"],
train_steps=params["NUM_STEPS"],
sample_1_of_n_eval_examples=params["SAMPLE_1_OF_N_EVAL_EXAMPLES"],
sample_1_of_n_eval_on_train_examples=(
params["SAMPLE_1_OF_N_TRAIN_EXAMPLES"]))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if params["CHECKPOINT_DIR"]:
if params["EVAL_TRAIN_DATA"]:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if params["RUN_ONCE"]:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
params["CHECKPOINT_DIR"]))
else:
model_lib.continuous_eval(estimator, params["CHECKPOINT_DIR"],
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required("model_dir")
flags.mark_flag_as_required("pipeline_config_path")
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples
),
)
estimator = train_and_eval_dict["estimator"]
train_input_fn = train_and_eval_dict["train_input_fn"]
eval_input_fns = train_and_eval_dict["eval_input_fns"]
eval_on_train_input_fn = train_and_eval_dict["eval_on_train_input_fn"]
predict_input_fn = train_and_eval_dict["predict_input_fn"]
train_steps = train_and_eval_dict["train_steps"]
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = "training_data"
input_fn = eval_on_train_input_fn
else:
name = "validation_data"
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(
input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(FLAGS.checkpoint_dir),
)
else:
model_lib.continuous_eval(
estimator,
FLAGS.checkpoint_dir,
input_fn,
train_steps,
name,
FLAGS.max_eval_retries,
)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False,
)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
tpu_cluster_resolver = (
tf.contrib.cluster_resolver.TPUClusterResolver(
tpu=[FLAGS.tpu_name],
zone=FLAGS.tpu_zone,
project=FLAGS.gcp_project))
tpu_grpc_url = tpu_cluster_resolver.get_master()
config = tf.contrib.tpu.RunConfig(
master=tpu_grpc_url,
evaluation_master=tpu_grpc_url,
model_dir=FLAGS.model_dir,
tpu_config=tf.contrib.tpu.TPUConfig(
iterations_per_loop=FLAGS.iterations_per_loop,
num_shards=FLAGS.num_shards))
kwargs = {}
if FLAGS.train_batch_size:
kwargs['batch_size'] = FLAGS.train_batch_size
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples),
use_tpu_estimator=True,
use_tpu=FLAGS.use_tpu,
num_shards=FLAGS.num_shards,
save_final_config=FLAGS.mode == 'train',
**kwargs)
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.mode == 'train':
estimator.train(input_fn=train_input_fn, max_steps=train_steps)
# Continuously evaluating.
if FLAGS.mode == 'eval':
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# Currently only a single eval input is allowed.
input_fn = eval_input_fns[0]
model_lib.continuous_eval(estimator, FLAGS.model_dir, input_fn, train_steps,
name)
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
tpu_cluster_resolver = (
tf.contrib.cluster_resolver.TPUClusterResolver(
tpu=[FLAGS.tpu_name],
zone=FLAGS.tpu_zone,
project=FLAGS.gcp_project))
tpu_grpc_url = tpu_cluster_resolver.get_master()
config = tf.contrib.tpu.RunConfig(
master=tpu_grpc_url,
evaluation_master=tpu_grpc_url,
model_dir=FLAGS.model_dir,
tpu_config=tf.contrib.tpu.TPUConfig(
iterations_per_loop=FLAGS.iterations_per_loop,
num_shards=FLAGS.num_shards))
kwargs = {}
if FLAGS.train_batch_size:
kwargs['batch_size'] = FLAGS.train_batch_size
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples),
use_tpu_estimator=True,
use_tpu=FLAGS.use_tpu,
num_shards=FLAGS.num_shards,
save_final_config=FLAGS.mode == 'train',
**kwargs)
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.mode == 'train':
estimator.train(input_fn=train_input_fn, max_steps=train_steps)
# Continuously evaluating.
if FLAGS.mode == 'eval':
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# Currently only a single eval input is allowed.
input_fn = eval_input_fns[0]
model_lib.continuous_eval(estimator, FLAGS.model_dir, input_fn, train_steps,
name)
def train_and_eval(rpn_type=None,
filter_fn_arg=None,
replace_rpn_arg=None,
number_of_stages=None):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir,
save_checkpoints_steps=10000)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples),
rpn_type=rpn_type,
filter_fn_arg=filter_fn_arg,
replace_rpn_arg=replace_rpn_arg,
number_of_stages=number_of_stages)
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
pass
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
config = tf.estimator.RunConfig(
model_dir=FLAGS.model_dir,
keep_checkpoint_max=FLAGS.keep_checkpoint_max,
save_checkpoints_steps=FLAGS.save_checkpoints_steps,
log_step_count_steps=FLAGS.log_step_count_steps)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name,
FLAGS.max_eval_retries)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
num_eval_steps=FLAGS.num_eval_steps,
eval_throttle_secs=FLAGS.eval_throttle_secs,
eval_start_delay_secs=FLAGS.eval_start_delay_secs,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
eval_steps=FLAGS.num_eval_steps)
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fn = train_and_eval_dict['eval_input_fn']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
eval_steps = train_and_eval_dict['eval_steps']
print('train_and_eval_dict: ' + str(train_and_eval_dict))
if FLAGS.checkpoint_dir:
print('enter into "FLAGS.checkpoint_dir:"')
if FLAGS.eval_training_data:
print('enter into "if FLAGS.eval_training_data:"')
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
print('enter into else of "if FLAGS.eval_training_data:"')
name = 'validation_data'
input_fn = eval_input_fn
if FLAGS.run_once:
print('enter into "if FLAGS.run_once:"')
estimator.evaluate(input_fn,
eval_steps,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
print('enter into else of "if FLAGS.run_once:"')
model_lib.continuous_eval(estimator, FLAGS.model_dir, input_fn,
eval_steps, train_steps, name)
else:
print('enter into else of "FLAGS.checkpoint_dir:"')
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fn,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.logging.set_verbosity(tf.logging.INFO)
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
tf.logging.set_verbosity(tf.logging.INFO)
#session_config = tf.ConfigProto()
#session_config.gpu_options.per_process_gpu_memory_fraction = 0.9
#session_config.gpu_options.allow_growth = True
#config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir, session_config=session_config)
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
num_eval_steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
#There are two arguments that must be provided.
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
#The variable 'config' contains instructions on saving checkpoints and
#general training info.
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir,
log_step_count_steps=10,
save_checkpoints_steps=300)
# the 'create_estimator_and_inputs' function
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(
FLAGS.hparams_overrides), #this is
# worth paying attention to, considering that I will need to do some tuning.
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
pipeline_config_path = "D:\\project3_faster_rcnn\\models-master\\research\\hat_dataset\\hat_resnet50_config.config"
model_dir = "D:\\project3_faster_rcnn\\models-master\\research\\hat_dataset\\checkpoints\\"
config = tf.estimator.RunConfig(model_dir=model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
config = tf.ConfigProto(gpu_options=tf.GPUOptions(
per_process_gpu_memory_fraction=0.9))
config.gpu_options.allow_growth = True
session = tf.Session(config=config)
set_session(session)
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
num_eval_steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required('train_dir')
flags.mark_flag_as_required('pipeline_config_path')
#config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
config = tf.estimator.RunConfig(
'path where the training output files will be generated')
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
#pipeline_config_path=FLAGS.pipeline_config_path,
pipeline_config_path='path of the model config file',
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
#if ('C:/Users/zansh/Anaconda3/Lib/site-packages/object_detection1/training/'):
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
num_eval_steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(_):
"""
Main program.
:param _: unused parameter.
:return: void.
"""
config = tf.estimator.RunConfig(model_dir=TRAINER_CONFIG["checkpoints_dir"])
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(TRAINER_CONFIG["hparams_override"]),
pipeline_config_path=TRAINER_CONFIG["pipeline_config_path"],
train_steps=TRAINER_CONFIG["num_train_steps"],
sample_1_of_n_eval_examples=TRAINER_CONFIG["sample_1_of_n_eval_examples"],
sample_1_of_n_eval_on_train_examples=(TRAINER_CONFIG["sample_1_of_n_eval_on_train_example"]))
estimator = train_and_eval_dict["estimator"]
train_input_fn = train_and_eval_dict["train_input_fn"]
eval_input_fns = train_and_eval_dict["eval_input_fns"]
eval_on_train_input_fn = train_and_eval_dict["eval_on_train_input_fn"]
predict_input_fn = train_and_eval_dict["predict_input_fn"]
train_steps = train_and_eval_dict["train_steps"]
# Train model from checkpoints.
if args.checkpoints and len(os.listdir(TRAINER_CONFIG["checkpoints_dir"])) > 1:
if TRAINER_CONFIG["eval_training_data"]:
name = "training_data"
input_fn = eval_on_train_input_fn
else:
name = "validation_data"
input_fn = eval_input_fns[0]
if args.run_once:
estimator.evaluate(input_fn, steps=None,
checkpoint_path=tf.train.latest_checkpoint(TRAINER_CONFIG["checkpoints_dir"]))
else:
model_lib.continuous_eval(estimator, TRAINER_CONFIG["checkpoints_dir"], input_fn,
train_steps, name)
# Train model from scratch.
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=TRAINER_CONFIG["eval_training_data"])
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
#flags.mark_flag_as_required('model_dir')
#flags.mark_flag_as_required('pipeline_config_path')
if not FLAGS.model_dir:
raise ValueError('You must supply the mode_dir')
if not FLAGS.pipeline_config_path:
raise ValueError('You must supply the pipeline_config_path')
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
eval_steps=FLAGS.num_eval_steps)
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fn = train_and_eval_dict['eval_input_fn']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
eval_steps = train_and_eval_dict['eval_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
input_fn = eval_input_fn
if FLAGS.run_once:
estimator.evaluate(input_fn,
eval_steps,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.model_dir, input_fn,
eval_steps, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fn,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
tpu_cluster_resolver = (
tf.contrib.cluster_resolver.python.training.TPUClusterResolver(
tpu_names=[FLAGS.tpu_name],
zone=FLAGS.tpu_zone,
project=FLAGS.gcp_project))
tpu_grpc_url = tpu_cluster_resolver.get_master()
config = tpu_config.RunConfig(
master=tpu_grpc_url,
evaluation_master=tpu_grpc_url,
model_dir=FLAGS.model_dir,
tpu_config=tpu_config.TPUConfig(
iterations_per_loop=FLAGS.iterations_per_loop,
num_shards=FLAGS.num_shards))
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
eval_steps=FLAGS.num_eval_steps,
use_tpu_estimator=True,
use_tpu=FLAGS.use_tpu,
num_shards=FLAGS.num_shards,
batch_size=FLAGS.train_batch_size)
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fn = train_and_eval_dict['eval_input_fn']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
train_steps = train_and_eval_dict['train_steps']
eval_steps = train_and_eval_dict['eval_steps']
if FLAGS.mode == 'train':
estimator.train(input_fn=train_input_fn, max_steps=train_steps)
# Continuously evaluating.
if FLAGS.mode == 'eval':
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
input_fn = eval_input_fn
model_lib.continuous_eval(estimator, FLAGS.model_dir, input_fn,
eval_steps, train_steps, name)
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
# strategy = tf.contrib.distribute.MirroredStrategy(num_gpus=4) # for multi-gpu
# config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir, train_distribute=strategy) # for multi-gpu
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
configs = config_util.get_configs_from_pipeline_file(FLAGS.pipeline_config_path)
configs['model'].ssd.freeze_batchnorm = FLAGS.freeze_batchnorm
configs['model'].ssd.num_classes = FLAGS.num_classes
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
pipeline_config_path=configs,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir, input_fn,
train_steps, name, FLAGS.max_eval_retries)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(_):
with open('system_dict.json') as json_file:
args = json.load(json_file)
config = tf.estimator.RunConfig(model_dir=args["model_dir"])
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
pipeline_config_path=args["pipeline_config_path"],
train_steps=args["num_train_steps"],
sample_1_of_n_eval_examples=args["sample_1_of_n_eval_examples"],
sample_1_of_n_eval_on_train_examples=(
args["sample_1_of_n_eval_on_train_examples"]))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if args["checkpoint_dir"]:
if args["eval_training_data"]:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if args["run_once"]:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
args["checkpoint_dir"]))
else:
model_lib.continuous_eval(estimator, args["checkpoint_dir"],
input_fn, train_steps, name,
args["max_eval_retries"])
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir, input_fn,
train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(_):
model_dir = os.path.join(os.getcwd(), FLAGS.model_dir)
pipeline_config_path = os.path.join(os.getcwd(), FLAGS.pipeline_config_path)
num_train_steps = FLAGS.num_train_steps
config = tf.estimator.RunConfig(model_dir=model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=pipeline_config_path,
train_steps=num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=FLAGS.sample_1_of_n_eval_on_train_examples)
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
num_eval_steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir, input_fn,
train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
config = tf.estimator.RunConfig(model_dir=model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(hparams_overrides),
pipeline_config_path=pipeline_config_path,
train_steps=num_train_steps,
sample_1_of_n_eval_examples=sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if checkpoint_dir:
if eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
## The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if run_once:
estimator.evaluate(
input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(checkpoint_dir))
else:
model_lib.continuous_eval(estimator, checkpoint_dir, input_fn,
train_steps, name, max_eval_retries)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
## Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
#config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
# Get info about full dataset
dataset,videos = get_dataset(data_info)
# Get experiment information from FLAGS
name = FLAGS.name
num_cycles = int(FLAGS.cycles)
run_num = int(FLAGS.run)
#num_steps = str(train_config.num_steps)
epochs = int(FLAGS.epochs)
restart_cycle = int(FLAGS.restart_from_cycle)
if FLAGS.checkpoint_dir:
"""
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
num_eval_steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir, input_fn,
train_steps, name)
"""
else:
# Load active set from cycle 0 and point to right model
if restart_cycle==0:
model_dir = FLAGS.model_dir + 'R' + str(run_num) + 'cycle0/'
#train_config.fine_tune_checkpoint = model_dir + 'model.ckpt'
else:
model_dir = FLAGS.model_dir + name + 'R' + str(run_num) + 'cycle' + str(restart_cycle) + '/'
# Get actual checkpoint model
#with open(model_dir+'checkpoint','r') as cfile:
# line = cfile.readlines()
# train_config.fine_tune_checkpoint = line[0].split(' ')[1][1:-2]
active_set = []
unlabeled_set=[]
with open(model_dir + 'active_set.txt', 'r') as f:
for line in f:
active_set.append(int(line))
for cycle in range(restart_cycle+1,num_cycles+1):
#### Evaluation of trained model on unlabeled set to obtain data for selection
if 'Rnd' not in name and cycle < num_cycles:
eval_train_dir = model_dir + name + 'R' + str(run_num) + 'cycle' + str(cycle) + 'eval_train/'
if os.path.exists(eval_train_dir + 'detections.dat'):
with open(eval_train_dir + 'detections.dat','rb') as infile:
###### pdb remove latinq
detected_boxes = pickle.load(infile)
#detected_boxes = pickle.load(infile,encoding='latin1')
else:
# Get unlabeled set
data_info['output_path'] = FLAGS.data_dir + 'AL/tfrecords/' + name + 'R' + str(run_num) + 'cycle' + str(cycle) + '_unlabeled.record'
# Do not evaluate labeled samples, their neighbors or unverified frames
aug_active_set = sel.augment_active_set(dataset,videos,active_set,num_neighbors=5)
unlabeled_set = [f['idx'] for f in dataset if f['idx'] not in aug_active_set and f['verified']]
# For TCFP, we need to get detections for pretty much every frame,
# as not candidates can may be used to support candidates
if ('TCFP' in name):
unlabeled_set = [i for i in range(len(dataset))]
print('Unlabeled frames in the dataset: {}'.format(len(unlabeled_set)))
save_tf_record(data_info,unlabeled_set)
"""
configs = config_util.get_configs_from_pipeline_file(pipeline_config_path=FLAGS.pipeline_config_path,
config_override=None)
eval_input_configs = configs['eval_input_configs']
eval_config = configs['eval_config']
model_config = configs['model']
eval_input_configs = configs['eval_input_configs']
MODEL_BUILD_UTIL_MAP = {'create_eval_input_fn': inputs.create_eval_input_fn}
create_eval_input_fn = MODEL_BUILD_UTIL_MAP['create_eval_input_fn']
eval_input_fns = [create_eval_input_fn(
eval_config=eval_config,
eval_input_config=eval_input_config,
model_config=model_config) for eval_input_config in eval_input_configs]
"""
# Set number of eval images to number of unlabeled samples and point to tfrecord
#eval_input_config.tf_record_input_reader.input_path[0] = data_info['output_path']
#eval_config.num_examples = len(unlabeled_set)
model_dir = FLAGS.model_dir + name + 'R' + str(run_num) + 'cycle' + str(cycle) + '/'
config = tf.estimator.RunConfig(model_dir=model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
Unlabeled_set_length=len(unlabeled_set),
Active_set_length=len(active_set),
epochs=epochs,
data_info=data_info,
FLAGS=FLAGS,
restart_cycle=restart_cycle,
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
"""
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
def get_next_eval_train(config):
return dataset_builder.make_initializable_iterator(
dataset_builder.build(config)).get_next()
# Initialize input dict again (necessary?)
#create_eval_train_input_dict_fn = functools.partial(get_next_eval_train, eval_input_config)
graph_rewriter_fn = None
if 'graph_rewriter_config' in configs:
graph_rewriter_fn = graph_rewriter_builder.build(
configs['graph_rewriter_config'], is_training=False)
# Need to reset graph for evaluation
tf.reset_default_graph()
#if FLAGS.eval_training_data:
#name = 'evaluation_of_training_data'
#input_fn = eval_on_train_input_fn
#else:
# name = 'validation_data'
# # The first eval input will be evaluated.
"""
input_fn = eval_input_fns[0]
if FLAGS.run_once:
predictions=estimator.evaluate(input_fn,
checkpoint_path=tf.train.latest_checkpoint(eval_train_dir))
else:
pdb.set_trace()
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir, input_fn,
train_steps, name)
pdb.set_trace()
#visualize_detections(dataset, unlabeled_set, detected_boxes, groundtruth_boxes)
with open(eval_train_dir + 'detections.dat','wb') as outfile:
pickle.dump(detected_boxes,outfile, protocol=pickle.HIGHEST_PROTOCOL)
print('Done computing detections in training set')
# Remove tfrecord used for training
if os.path.exists(data_info['output_path']):
os.remove(data_info['output_path'])
#### Training of current cycle
model_dir = FLAGS.model_dir + name + 'R' + str(run_num) + 'cycle' + str(cycle) + '/'
config = tf.estimator.RunConfig(model_dir=model_dir)
# Budget for each cycle is the number of videos (0.5% of train set)
if ('Rnd' in name):
#indices = select_random_video(dataset,videos,active_set)
#indices = sel.select_random(dataset,videos,active_set,budget=num_videos)
indices = sel.select_random_video(dataset,videos,active_set)
else:
if ('Ent' in name):
indices = sel.select_entropy_video(dataset,videos,FLAGS.data_dir,active_set,detected_boxes)
elif ('Lst' in name):
indices = sel.select_least_confident_video(dataset,videos,active_set,detected_boxes)
elif ('TCFP' in name):
indices = sel.select_TCFP_per_video(dataset,videos,FLAGS.data_dir,active_set,detected_boxes)
elif ('FP_gt' in name):
indices = sel.selectFpPerVideo(dataset,videos,active_set,detected_boxes,groundtruth_boxes,cycle)
elif ('FN_gt' in name):
indices = sel.selectFnPerVideo(dataset,videos,active_set,detected_boxes,groundtruth_boxes,cycle)
elif ('FPN' in name):
indices = sel.select_FPN_PerVideo(dataset,videos,active_set,detected_boxes,groundtruth_boxes,cycle)
active_set.extend(indices)
data_info['output_path'] = FLAGS.data_dir + 'AL/tfrecords/' + name + 'R' + str(run_num) + 'cycle' + str(cycle) + '.record'
save_tf_record(data_info,active_set)
pdb.set_trace()
train_and_eval_dict = model_lib.create_estimator_and_inputs(
Unlabeled_set_length=len(unlabeled_set),
Active_set_length=len(active_set),
epochs=epochs,
data_info=data_info,
FLAGS=FLAGS,
restart_cycle=restart_cycle,
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
print('-----------------train and evaluation-------------------------')
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
#Save active_set in train dir in case we want to restart training
with open(model_dir + 'active_set.txt', 'w') as f:
for item in active_set:
f.write('{}\n'.format(item))
# Remove tfrecord used for training
if os.path.exists(data_info['output_path']):
os.remove(data_info['output_path'])
def main(unused_argv):
print("*** train.py/main()")
# flags.mark_flag_as_required('model_dir')
# flags.mark_flag_as_required('pipeline_config_path')
print('*** FLAGS ***')
print("pipeline_config_path:", FLAGS.pipeline_config_path)
## --verification - debug
print("config exists:", os.path.exists(FLAGS.pipeline_config_path))
dir_list = [f for f in listdir(".")]
for item in dir_list:
print("file:", item)
print("model_dir:", FLAGS.model_dir)
print("train:", FLAGS.train)
print("val:", FLAGS.val)
print("sample_1_of_n_eval_examples:", FLAGS.sample_1_of_n_eval_examples)
print("hparams_overrides:", FLAGS.hparams_overrides)
print("checkpoint_dir:", FLAGS.checkpoint_dir)
# check pipeline config pararameters
# - input data
pipeline_config_dict = config_util.get_configs_from_pipeline_file(
FLAGS.pipeline_config_path)
check_input_data_existance(pipeline_config_dict)
print(" - - - - - - - - -")
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
tf.enable_eager_execution()
tf.set_random_seed(0)
tf.logging.set_verbosity(tf.logging.ERROR)
# Creates `Estimator`, input functions, and steps
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
# so here are the outputs (that were in a dict)
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
tf.logging.set_verbosity(tf.logging.INFO)
if FLAGS.amp:
os.environ["TF_ENABLE_AUTO_MIXED_PRECISION"] = "1"
else:
os.environ["TF_ENABLE_AUTO_MIXED_PRECISION"] = "0"
# Set seed to reduce randomness
np.random.seed(FLAGS.seed)
tf.set_random_seed(FLAGS.seed)
hvd.init()
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
session_config = tf.ConfigProto()
session_config.gpu_options.per_process_gpu_memory_fraction=0.9
session_config.gpu_options.visible_device_list = str(hvd.local_rank())
if FLAGS.allow_xla:
session_config.graph_options.optimizer_options.global_jit_level = tf.OptimizerOptions.ON_1
model_dir = FLAGS.model_dir if hvd.rank() == 0 else None
config = tf.estimator.RunConfig(tf_random_seed=(FLAGS.seed + hvd.rank()),
model_dir=model_dir, session_config=session_config)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
eval_count=FLAGS.eval_count,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir, input_fn,
train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
train_hooks = [hvd.BroadcastGlobalVariablesHook(0), DLLoggerHook(hvd.size()*train_and_eval_dict['train_batch_size'], hvd.rank())]
eval_hooks = []
for x in range(FLAGS.eval_count):
estimator.train(train_input_fn,
hooks=train_hooks,
steps=train_steps // FLAGS.eval_count)
if hvd.rank() == 0 and not FLAGS.train_only:
eval_input_fn = eval_input_fns[0]
results = estimator.evaluate(eval_input_fn,
steps=None,
hooks=eval_hooks)
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
tpu_cluster_resolver = (tf.contrib.cluster_resolver.TPUClusterResolver(
tpu=FLAGS.tpu_name.split(','),
zone=FLAGS.tpu_zone,
project=FLAGS.gcp_project))
tpu_grpc_url = tpu_cluster_resolver.get_master()
config = tf.contrib.tpu.RunConfig(
master=tpu_grpc_url,
evaluation_master=tpu_grpc_url,
model_dir=FLAGS.model_dir,
tpu_config=tf.contrib.tpu.TPUConfig(
iterations_per_loop=FLAGS.iterations_per_loop,
num_shards=FLAGS.num_shards),
save_checkpoints_steps=FLAGS.eval_every_n_iters)
kwargs = {}
if FLAGS.train_batch_size:
kwargs['batch_size'] = FLAGS.train_batch_size
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples),
use_tpu_estimator=True,
use_tpu=FLAGS.use_tpu,
num_shards=FLAGS.num_shards,
**kwargs)
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.mode == 'train':
estimator.train(input_fn=train_input_fn, max_steps=train_steps)
# Continuously evaluating.
if FLAGS.mode == 'eval':
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# Currently only a single eval input is allowed.
input_fn = eval_input_fns[0]
model_lib.continuous_eval(estimator, FLAGS.model_dir, input_fn,
train_steps, name)
if FLAGS.mode == 'train_and_eval':
# load last checkpoint and start from there
current_step = load_global_step_from_checkpoint_dir(FLAGS.model_dir)
while current_step < train_steps:
# Train for up to steps_per_eval number of steps.
# At the end of training, a checkpoint will be written to --model_dir.
next_checkpoint = min(current_step + FLAGS.eval_every_n_iters,
train_steps)
estimator.train(input_fn=train_input_fn, max_steps=next_checkpoint)
current_step = next_checkpoint
# Evaluate the model on the most recent model in --model_dir.
# Since evaluation happens in batches of --eval_batch_size, some images
# may be excluded modulo the batch size. As long as the batch size is
# consistent, the evaluated images are also consistent.
tf.logging.info('Starting to evaluate at step %d', next_checkpoint)
# Background evaluation process.
args = [
'PYTHONPATH=$PYTHONPATH:slim', 'python',
'object_detection/model_main.py', '--pipeline_config_path',
FLAGS.pipeline_config_path, '--model_dir', FLAGS.model_dir,
'--checkpoint_dir', FLAGS.model_dir,
'--sample_1_of_n_eval_examples',
str(FLAGS.sample_1_of_n_eval_examples),
'--gpu_memory_fraction',
str(0.2), '--run_once',
str(True)
]
print(' '.join(args))
p = subprocess.Popen(' '.join(args),
stderr=subprocess.STDOUT,
shell=True)
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
# config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
# config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir,
# save_checkpoints_steps=5000)
config = tf.estimator.RunConfig(
model_dir=FLAGS.model_dir,
#save_checkpoints_steps=5000,
save_checkpoints_steps=1000,
save_checkpoints_secs=None,
keep_checkpoint_max=
None, # useless: model_lib.py tf.train.Saver-> max_to_keep take effect
keep_checkpoint_every_n_hours=10000)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
# import pdb
# pdb.set_trace()
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
# throttle_secs is not documented in eval.proto. This replaces eval_interval_secs somewhat
throttle_secs = 60 * 60 # every 60 min
eval_spec = eval_specs[0]
my_eval_spec = tf.estimator.EvalSpec(
name=eval_spec.name,
input_fn=eval_spec.input_fn,
steps=None,
exporters=eval_spec.exporters,
start_delay_secs=
1800, # 30 minutes - does not seem to be respected...
throttle_secs=throttle_secs)
print('=========== my_eval_spec')
print(my_eval_spec)
print('=========================')
tf.estimator.train_and_evaluate(estimator, train_spec, my_eval_spec)
def main(unused_argv):
tf.logging.set_verbosity(tf.logging.INFO)
hvd.init()
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
session_config = tf.ConfigProto()
session_config.gpu_options.allow_growth = True
session_config.gpu_options.visible_device_list = str(hvd.local_rank())
if FLAGS.allow_xla:
session_config.graph_options.optimizer_options.global_jit_level = tf.OptimizerOptions.ON_1
model_dir = FLAGS.model_dir if hvd.rank() == 0 else None
config = tf.estimator.RunConfig(model_dir=model_dir,
session_config=session_config)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
eval_count=FLAGS.eval_count,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
train_hooks = [hvd.BroadcastGlobalVariablesHook(0)]
eval_hooks = []
for x in range(FLAGS.eval_count):
estimator.train(train_input_fn,
hooks=train_hooks,
steps=train_steps // FLAGS.eval_count)
if hvd.rank() == 0:
eval_input_fn = eval_input_fns[0]
results = estimator.evaluate(eval_input_fn,
steps=None,
hooks=eval_hooks)
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
if FLAGS.gpu_device is not None:
os.environ["CUDA_VISIBLE_DEVICES"] = str(FLAGS.gpu_device)
session_config = tf.ConfigProto()
session_config.gpu_options.allow_growth = True
config = tf.estimator.RunConfig(
model_dir=FLAGS.model_dir,
session_config=session_config,
save_checkpoints_secs=FLAGS.save_checkpoints_secs)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name,
FLAGS.max_eval_retries)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Multiple Eval Specs allowed.
# TODO: Fix name of saving_listeners
saving_listeners = [
EvalCheckpointSaverListener(estimator, eval_specs[0].input_fn,
'validation')
]
if len(eval_specs) > 1:
saving_listeners.append(
EvalCheckpointSaverListener(estimator, eval_specs[1].input_fn,
'training'))
estimator.train(input_fn=train_spec.input_fn,
max_steps=train_spec.max_steps,
saving_listeners=saving_listeners)
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
# Soft placement allows placing on CPU ops without GPU implementation.
session_config = tf.ConfigProto(allow_soft_placement=True,
log_device_placement=False)
session_config.gpu_options.per_process_gpu_memory_fraction = 0.95
session_config.gpu_options.visible_device_list = "1"
log_step_cnt = 10
save_checkpoints_steps = 25000
tf.logging.set_verbosity(tf.logging.INFO)
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir,
save_checkpoints_steps=save_checkpoints_steps,
session_config=session_config,
log_step_count_steps=log_step_cnt)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
#num_eval_steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir, input_fn,
train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
current_time_start = datetime.now(tz_SG).strftime('%d-%m-%Y %H:%M:%S')
start=time.time()
log_directory = os.path.join(os.getcwd(),FLAGS.model_dir)
#make directory for log files
if not os.path.exists(log_directory):
os.makedirs(log_directory)
print("Logging will be found in {}".format(log_directory))
log_file = os.path.join(log_directory, 'log.txt')
# create file handler which logs event debug messages
log = logging.getLogger('tensorflow')
log.root.handlers[0].setFormatter(CsvFormatter(output_csv = os.path.join(log_directory, 'log.csv')))
#log.disable(logging.WARNING)
log.addFilter(StepLossFilter())
config2 = tf.ConfigProto()
config2.gpu_options.allow_growth = True
log.setLevel(logging.INFO)
#formatter = logging.Formatter('%(levelname)s - %(message)s')
formatter = logging.Formatter('%(message)s')
#FileHandler is used to send the log entries to a file
fh = logging.FileHandler(log_file)
print("File handler: {}".format(fh))
fh.setLevel(logging.INFO)
fh.setFormatter(formatter)
log.addHandler(fh)
# StreamHandler is used to send the log entries to console
ch = logging.StreamHandler()
ch.addFilter(StepLossFilter())
ch.setLevel(logging.INFO)
ch.setFormatter(formatter)
log.addHandler(ch)
#Log the estimator steps
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir,save_checkpoints_steps=500, log_step_count_steps=100,session_config=config2)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir, input_fn,
train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
tf.estimator.train_and_evaluate(estimator, train_spec, eval_specs[0])
end=time.time()
current_time_end = datetime.now(tz_SG).strftime('%d-%m-%Y %H:%M:%S')
log.info("Started: {}".format(current_time_start))
log.info("Ended: {}".format(current_time_end))
log.info("Duration: {} secs".format(round(end-start,0)))
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
# substitute the AML data reference mount points for relevant parts in the pipeline.config and overwrite
with open(FLAGS.pipeline_config_path) as f:
config_file = f.read()
new_config_file = config_file.replace(
'$AZUREML_DATAREFERENCE_tfrecords',
actual_path_tfrecords).replace('$AZUREML_DATAREFERENCE_artifacts',
actual_path_artifacts)
with open(FLAGS.pipeline_config_path, 'w') as f:
f.write(new_config_file)
print('model_main.py, main(), finished substituting mount points.')
config = tf.estimator.RunConfig(
model_dir=FLAGS.model_dir,
save_checkpoints_steps=
104012 # save less often than default - 1/5 of an epoch
)
train_and_eval_dict = model_lib.create_estimator_and_inputs(
run_config=config,
hparams=model_hparams.create_hparams(FLAGS.hparams_overrides),
pipeline_config_path=FLAGS.pipeline_config_path,
train_steps=FLAGS.num_train_steps,
sample_1_of_n_eval_examples=FLAGS.sample_1_of_n_eval_examples,
sample_1_of_n_eval_on_train_examples=(
FLAGS.sample_1_of_n_eval_on_train_examples))
estimator = train_and_eval_dict['estimator']
train_input_fn = train_and_eval_dict['train_input_fn']
eval_input_fns = train_and_eval_dict['eval_input_fns']
eval_on_train_input_fn = train_and_eval_dict['eval_on_train_input_fn']
predict_input_fn = train_and_eval_dict['predict_input_fn']
train_steps = train_and_eval_dict['train_steps']
if FLAGS.checkpoint_dir:
if FLAGS.eval_training_data:
name = 'training_data'
input_fn = eval_on_train_input_fn
else:
name = 'validation_data'
# The first eval input will be evaluated.
input_fn = eval_input_fns[0]
if FLAGS.run_once:
estimator.evaluate(input_fn,
steps=None,
checkpoint_path=tf.train.latest_checkpoint(
FLAGS.checkpoint_dir))
else:
model_lib.continuous_eval(estimator, FLAGS.checkpoint_dir,
input_fn, train_steps, name)
else:
train_spec, eval_specs = model_lib.create_train_and_eval_specs(
train_input_fn,
eval_input_fns,
eval_on_train_input_fn,
predict_input_fn,
train_steps,
eval_on_train_data=False)
# Currently only a single Eval Spec is allowed.
# throttle_secs is not documented in eval.proto. This replaces eval_interval_secs somewhat
throttle_secs = 60 * 60 # every 60 min
eval_spec = eval_specs[0]
my_eval_spec = tf.estimator.EvalSpec(
name=eval_spec.name,
input_fn=eval_spec.input_fn,
steps=None,
exporters=eval_spec.exporters,
start_delay_secs=
1800, # 30 minutes - does not seem to be respected...
throttle_secs=throttle_secs)
print('=========== my_eval_spec')
print(my_eval_spec)
print('=========================')
tf.estimator.train_and_evaluate(estimator, train_spec, my_eval_spec)