tenssor flow estimatorがhookを使ってfinetune方式を実現します。
6532 ワード
finetuneを実現するためには、次の2つの解決策があります。
モデルfnの中でモデルを定義してから直接に値を付けます。
tf.com ntrib.learn.Experimentを定義する際にtrin_を通過することができます。monitorsパラメータ指定
でも、個人的には一番いいと思います。やはりestimatorで定義して、experimentを実験のモード(訓練回数、検証回数など)だけに集中させます。
現在のモード:
mode:A ModeKeys.Specifris is trining,evalution or predition.
計算図
predictions:Precions Tensor dict of Tensor.
loss:Training loss Tensor.Must be either scalar,or with shape[1]。
トレイop:Op for the trining step.
eval_metric_ops:Dict of metric result keyed by name.The values of the dict are the result of carrling a metric function,namelya(metric_)テナンタ、udate_op)tuple.metric_tenssor shoud be evaluated without any impact on state(typically is a put computation result based on variables.).For example,it shoud not trigger the up dateop or requires any input feting.
エクスポートポリシー
export.outputts:Describes the output signature s to be export to
SavedModel and used during serving.A dict{name:output}where:
name:An arbitrary name for this output.
out put:an ExportOutput object such as Class ification Output,Regression Output,or PredictOutput.Single-headed models need to speciful inctry.Multi-headed modelstreconstants.DEFAULT_SERVING_SIGN ATURE_DEF_KEY.
chiefフックトレーニング時のモデル保存術フックCheckpointSaverHook、モデル回復など
trining_ちぇf_hooks:Iterable of tf.trin.Session RunHook to run on the chief worker during trining.
ウォーカーフックトレーニング時の監視戦略フックは、NanTensorHookロギングTensorhookなどです。
trining_hooks:Iterable of tf.trin.SessionRunHook to run on all works during trining.
初期化とsaverを指定します。
scaffold:A tf.trin.Scaffold object that can be used to set initiazation,saver,and more to be used in trining.
evalutionフック
evalution_hooks:Iterable of tf.trin.SessionRunHook to run during evalution.
カスタムフックは以下の通りです。
モデルfnの中でモデルを定義してから直接に値を付けます。
def model_fn(features, labels, mode, params):
# .....
# finetune
if params.checkpoint_path and (not tf.train.latest_checkpoint(params.model_dir)):
checkpoint_path = None
if tf.gfile.IsDirectory(params.checkpoint_path):
checkpoint_path = tf.train.latest_checkpoint(params.checkpoint_path)
else:
checkpoint_path = params.checkpoint_path
tf.train.init_from_checkpoint(
ckpt_dir_or_file=checkpoint_path,
assignment_map={params.checkpoint_scope: params.checkpoint_scope} # 'OptimizeLoss/':'OptimizeLoss/'
)
フックを使う。tf.com ntrib.learn.Experimentを定義する際にtrin_を通過することができます。monitorsパラメータ指定
# Define the experiment
experiment = tf.contrib.learn.Experiment(
estimator=estimator, # Estimator
train_input_fn=train_input_fn, # First-class function
eval_input_fn=eval_input_fn, # First-class function
train_steps=params.train_steps, # Minibatch steps
min_eval_frequency=params.eval_min_frequency, # Eval frequency
# train_monitors=[], # Hooks for training
# eval_hooks=[eval_input_hook], # Hooks for evaluation
eval_steps=params.eval_steps # Use evaluation feeder until its empty
)
tf.estimart.EstimatoSpecを定義する際にtriningを通してもいいです。ちぇf_hooksパラメータ指定。でも、個人的には一番いいと思います。やはりestimatorで定義して、experimentを実験のモード(訓練回数、検証回数など)だけに集中させます。
def model_fn(features, labels, mode, params):
# ....
return tf.estimator.EstimatorSpec(
mode=mode,
predictions=predictions,
loss=loss,
train_op=train_op,
eval_metric_ops=eval_metric_ops,
# scaffold=get_scaffold(),
# training_chief_hooks=None
)
ここでは、以下のtf.estimart.EstimatoSpecの役割を説明します。オブジェクトはモデルの正方形面を記述する。含む:現在のモード:
mode:A ModeKeys.Specifris is trining,evalution or predition.
計算図
predictions:Precions Tensor dict of Tensor.
loss:Training loss Tensor.Must be either scalar,or with shape[1]。
トレイop:Op for the trining step.
eval_metric_ops:Dict of metric result keyed by name.The values of the dict are the result of carrling a metric function,namelya(metric_)テナンタ、udate_op)tuple.metric_tenssor shoud be evaluated without any impact on state(typically is a put computation result based on variables.).For example,it shoud not trigger the up dateop or requires any input feting.
エクスポートポリシー
export.outputts:Describes the output signature s to be export to
SavedModel and used during serving.A dict{name:output}where:
name:An arbitrary name for this output.
out put:an ExportOutput object such as Class ification Output,Regression Output,or PredictOutput.Single-headed models need to speciful inctry.Multi-headed modelstreconstants.DEFAULT_SERVING_SIGN ATURE_DEF_KEY.
chiefフックトレーニング時のモデル保存術フックCheckpointSaverHook、モデル回復など
trining_ちぇf_hooks:Iterable of tf.trin.Session RunHook to run on the chief worker during trining.
ウォーカーフックトレーニング時の監視戦略フックは、NanTensorHookロギングTensorhookなどです。
trining_hooks:Iterable of tf.trin.SessionRunHook to run on all works during trining.
初期化とsaverを指定します。
scaffold:A tf.trin.Scaffold object that can be used to set initiazation,saver,and more to be used in trining.
evalutionフック
evalution_hooks:Iterable of tf.trin.SessionRunHook to run during evalution.
カスタムフックは以下の通りです。
class RestoreCheckpointHook(tf.train.SessionRunHook):
def __init__(self,
checkpoint_path,
exclude_scope_patterns,
include_scope_patterns
):
tf.logging.info("Create RestoreCheckpointHook.")
#super(IteratorInitializerHook, self).__init__()
self.checkpoint_path = checkpoint_path
self.exclude_scope_patterns = None if (not exclude_scope_patterns) else exclude_scope_patterns.split(',')
self.include_scope_patterns = None if (not include_scope_patterns) else include_scope_patterns.split(',')
def begin(self):
# You can add ops to the graph here.
print('Before starting the session.')
# 1. Create saver
#exclusions = []
#if self.checkpoint_exclude_scopes:
# exclusions = [scope.strip()
# for scope in self.checkpoint_exclude_scopes.split(',')]
#
#variables_to_restore = []
#for var in slim.get_model_variables(): #tf.global_variables():
# excluded = False
# for exclusion in exclusions:
# if var.op.name.startswith(exclusion):
# excluded = True
# break
# if not excluded:
# variables_to_restore.append(var)
#inclusions
#[var for var in tf.trainable_variables() if var.op.name.startswith('InceptionResnetV1')]
variables_to_restore = tf.contrib.framework.filter_variables(
slim.get_model_variables(),
include_patterns=self.include_scope_patterns, # ['Conv'],
exclude_patterns=self.exclude_scope_patterns, # ['biases', 'Logits'],
# If True (default), performs re.search to find matches
# (i.e. pattern can match any substring of the variable name).
# If False, performs re.match (i.e. regexp should match from the beginning of the variable name).
reg_search = True
)
self.saver = tf.train.Saver(variables_to_restore)
def after_create_session(self, session, coord):
# When this is called, the graph is finalized and
# ops can no longer be added to the graph.
print('Session created.')
tf.logging.info('Fine-tuning from %s' % self.checkpoint_path)
self.saver.restore(session, os.path.expanduser(self.checkpoint_path))
tf.logging.info('End fineturn from %s' % self.checkpoint_path)
def before_run(self, run_context):
#print('Before calling session.run().')
return None #SessionRunArgs(self.your_tensor)
def after_run(self, run_context, run_values):
#print('Done running one step. The value of my tensor: %s', run_values.results)
#if you-need-to-stop-loop:
# run_context.request_stop()
pass
def end(self, session):
#print('Done with the session.')
pass
以上のこのtenssor flow estimatorはヤフーを使ってfinetune方式を実現します。つまり、小編集で皆さんに共有する内容です。参考にしていただければと思います。よろしくお願いします。