时间戳

首页 > 解决方案 > 可以在训练过程中使用 keras 实时获取输出层吗？

python - 可以在训练过程中使用 keras 实时获取输出层吗？

问题描述

我尝试在训练期间获得输出层。我正在尝试实时对模型进行 3D 可视化并使其具有交互性。我正在使用带有 tensorflow 2.0 和 python 3 的谷歌 colab。

这是我的代码：

进口

  from __future__ import absolute_import, division, print_function, unicode_literals
 try:
   # Use the %tensorflow_version magic if in colab.
     %tensorflow_version 2.x
 except Exception:
       pass

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt

import tensorflow_hub as hub
import tensorflow_datasets as tfds

from tensorflow.keras import datasets, layers, models

from tensorflow.keras import backend as K
from tensorflow.keras.backend import clear_session

from tensorflow.keras.callbacks import Callback as Callback

import logging
logger = tf.get_logger()
logger.setLevel(logging.ERROR)

获取数据

splits = tfds.Split.TRAIN.subsplit([70, 30])

(training_set, validation_set), dataset_info = tfds.load('tf_flowers',with_info=True, as_supervised=True, split=splits)
 for i, example in enumerate(training_set.take(5)):
        print('Image {} shape: {} label: {}'.format(i+1, example[0].shape, example[1]))

检查类和图像的数量

 num_classes = dataset_info.features['label'].num_classes

 num_training_examples = 0
 num_validation_examples = 0

 for example in training_set:
   num_training_examples += 1

 for example in validation_set:
   num_validation_examples += 1

 print('Total Number of Classes: {}'.format(num_classes))
 print('Total Number of Training Images: {}'.format(num_training_examples))
 print('Total Number of Validation Images: {} \n'.format(num_validation_examples))

开始

   IMAGE_RES = 299
   BATCH_SIZE = 32
def format_image(image, label):
   image = tf.image.resize(image, (IMAGE_RES, IMAGE_RES))/255.0
  return image, label

 (training_set, validation_set), dataset_info = tfds.load('tf_flowers', with_info=True, as_supervised=True, split=splits)
  train_batches = training_set.shuffle(num_training_examples//4).map(format_image).batch(BATCH_SIZE).prefetch(1)
    validation_batches = validation_set.map(format_image).batch(BATCH_SIZE).prefetch(1)

URL = "https://tfhub.dev/google/tf2-preview/inception_v3/feature_vector/4"
feature_extractor = hub.KerasLayer(URL,
  input_shape=(IMAGE_RES, IMAGE_RES, 3),
trainable=False)

model_inception = tf.keras.Sequential([
feature_extractor,
layers.Dense(num_classes, activation='softmax')
])

 model_inception.summary()

这是我在训练期间尝试获取输出层的自定义回调

    import datetime
 from keras.callbacks import Callback

class MyCustomCallback(tf.keras.callbacks.Callback):

  def on_train_batch_begin(self, batch, logs=None):
     print('Training: batch {} begins at {}'.format(batch, datetime.datetime.now().time()))

  def on_train_batch_end(self, batch, logs=None):
     for i in range(len(model_inception.layers)):
      inp = self.model.input                                    # input placeholder
      outputs = [layer.output for layer in self.model.layers]     # all layer outputs
      functors = [K.function([inp, K.learning_phase()], [out]) for out in outputs]    # evaluation functions
      input_shape = [1] + list(self.model.input_shape[1:])
      test = np.random.random(input_shape)
      layer_outs = [func([test, 1.]) for func in functors] 
      print('\n Training: batch {} ends at {}'.format( layer_outs , datetime.datetime.now().time()))

  def on_test_batch_begin(self, batch, logs=None):
    print('Evaluating: batch {} begins at {}'.format(batch, datetime.datetime.now().time()))

  def on_test_batch_end(self, batch, logs=None):
   # layer_output = get_3rd_layer_output(self.validation_data)[0]  
    print('Training: batch {} ends at {} with the output layer {}'.format(batch, datetime.datetime.now().time()))

 The problem is in callback of how i can get the output/input of each layer at the end of each batch

这是使用我的自定义回调进行模型编译和训练 

 model_inception.compile(
  optimizer='adam', 
  loss='sparse_categorical_crossentropy',
  metrics=['accuracy'])

 EPOCHS = 2

 history = model_inception.fit(train_batches,
                epochs=EPOCHS,
                steps_per_epoch=20,
                validation_data=validation_batches,callbacks=[MyCustomCallback()])

我尝试运行它时的当前错误

AttributeError                            Traceback (most recent call last)
<ipython-input-10-5909c67ba93f> in <module>()
      9                     epochs=EPOCHS,
     10                     steps_per_epoch=20,
---> 11                     validation_data=validation_batches,callbacks=[MyCustomCallback()])
     12 
     13 # #Testing

11 frames
/tensorflow-2.0.0/python3.6/tensorflow_core/python/eager/lift_to_graph.py in <listcomp>(.0)
 247   # Check that the initializer does not depend on any placeholders.
 248   sources = object_identity.ObjectIdentitySet(sources or [])
-->249   visited_ops = set([x.op for x in sources])
 250   op_outputs = collections.defaultdict(set)
 251 

AttributeError: 'int' object has no attribute 'op'

标签： pythontensorflowkerascallbackdeep-learning

解决方案

如果您阅读了自定义回调的来源，请点击此处

我们定义的每个自定义回调都有一个属性模型。

您可以在 Cutomcallbacks 中定义的函数中使用模型对象。

例如，

def on_train_batch_end(self, batch, logs=None):
    #here you can get the model reference. 
    self.model.predict(dummy_data)

self.model 是 keras.models.Model 的一个实例，可以使用它调用相应的函数。

更多参考可以在这里和这里找到

请按照以下评论获得答案。

[编辑 1]

OP评论中的代码段

def on_train_batch_end(self, batch, logs=None): 
    for i in range(len(model_inception.layers)): 
        get_layer_output = K.function(inputs = self.model.layers[i].input, outputs = self.model.layers[i].output) 
        print('\n Training: output of the layer {} is {} ends at {}'.format(i, get_layer_output.outputs , datetime.datetime.now().time()))

推荐阅读