个人中心
gpt4 book ai didi

tensorflow estimator 使用hook实现finetune方式

转载 作者：qq735679552 更新时间：2022-09-29 22:32:09 33 4
gpt4 key购买 nike

CFSDN坚持开源创造价值，我们致力于搭建一个资源共享平台，让每一个IT人在这里找到属于你的精彩世界.

这篇CFSDN的博客文章tensorflow estimator 使用hook实现finetune方式由作者收集整理，如果你对这篇文章有兴趣，记得点赞哟.

为了实现finetune有如下两种解决方案:

model_fn里面定义好模型之后直接赋值 。

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
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/'
)

使用钩子 hooks.

可以在定义tf.contrib.learn.Experiment的时候通过train_monitors参数指定 。

?
1
2
3
4
5
6
7
8
9
10
11
# 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
)

也可以在定义tf.estimator.EstimatorSpec 的时候通过training_chief_hooks参数指定.

不过个人觉得最好还是在estimator中定义，让experiment只专注于控制实验的模式（训练次数,验证次数等等）.

?
1
2
3
4
5
6
7
8
9
10
11
12
13
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
  )

这里顺便解释以下tf.estimator.EstimatorSpec对像的作用。该对象描述来一个模型的方方面面。包括:

当前的模式:

mode: A ModeKeys. Specifies if this is training, evaluation or prediction. 。

计算图 。

predictions: Predictions Tensor or dict of Tensor. 。

loss: Training loss Tensor. Must be either scalar, or with shape [1]. 。

train_op: Op for the training step. 。

eval_metric_ops: Dict of metric results keyed by name. The values of the dict are the results of calling a metric function, namely a (metric_tensor, update_op) tuple. metric_tensor should be evaluated without any impact on state (typically is a pure computation results based on variables.). For example, it should not trigger the update_op or requires any input fetching. 。

导出策略 。

export_outputs: Describes the output signatures to be exported to 。

SavedModel and used during serving. A dict {name: output} where

name: An arbitrary name for this output. 。

output: an ExportOutput object such as ClassificationOutput, RegressionOutput, or PredictOutput. Single-headed models only need to specify one entry in this dictionary. Multi-headed models should specify one entry for each head, one of which must be named using signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY. 。

chief钩子 训练时的模型保存策略钩子CheckpointSaverHook， 模型恢复等 。

training_chief_hooks: Iterable of tf.train.SessionRunHook objects to run on the chief worker during training. 。

worker钩子 训练时的监控策略钩子如： NanTensorHook LoggingTensorHook 等 。

training_hooks: Iterable of tf.train.SessionRunHook objects to run on all workers during training. 。

指定初始化和saver 。

scaffold: A tf.train.Scaffold object that can be used to set initialization, saver, and more to be used in training. 。

evaluation钩子 。

evaluation_hooks: Iterable of tf.train.SessionRunHook objects to run during evaluation. 。

自定义的钩子如下:

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
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

以上这篇tensorflow estimator 使用hook实现finetune方式就是小编分享给大家的全部内容了，希望能给大家一个参考，也希望大家多多支持我.

原文链接：https://blog.csdn.net/andylei777/article/details/79074757 。

最后此篇关于tensorflow estimator 使用hook实现finetune方式的文章就讲到这里了,如果你想了解更多关于tensorflow estimator 使用hook实现finetune方式的内容请搜索CFSDN的文章或继续浏览相关文章，希望大家以后支持我的博客！ 。

33 4 0
文章推荐： TFRecord格式存储数据与队列读取实例
文章推荐： Python实现FLV视频拼接功能
文章推荐： TensorFlow——Checkpoint为模型添加检查点的实例
文章推荐： 在tensorflow中设置保存checkpoint的最大数量实例
qq735679552
个人简介

我是一名优秀的程序员,十分优秀！

滴滴打车优惠券免费领取
滴滴打车优惠券
全站热门文章
Copyright 2021 - 2024 cfsdn All Rights Reserved 蜀ICP备2022000587号
广告合作:1813099741@qq.com 6ren.com