Question

我是TensorFlow的新手，我正在尝试使用Keras创建分类器。我的训练数据分为两个文件： -一个带有训练示例的示例，每个示例都是64个浮点数的向量 -第二个带有标签的标签，每个标签都是一个整数（0，..，SIZE）（SIZE为100），它描述了一个类。

两个文件都需要大文件，我无法将它们装入内存，因此我使用了tf.Dataset。我创建了两个数据集（一个用于功能，一个用于标签），然后使用tf.data.Dataset.zip（）将它们合并。但是，在训练过程中，出现“ IndexError：列表索引超出范围”错误。但是当我打印输入数据时，它看起来还不错。这是代码：

model = tf.keras.Sequential()
model.add(layers.Dense(256, activation='relu'))
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(SIZE, activation='softmax'))
opt = tf.keras.optimizers.Adagrad()
model.compile(optimizer=opt, 
              loss='categorical_crossentropy',
              metrics=['accuracy'])

#read data to: data and label
dataset = tf.data.Dataset.zip((data, label))
#iterator = dataset.make_one_shot_iterator()
#next_element = iterator.get_next()
#print(next_element[0])
#print("\n\n")
#print(next_element[1])

model.fit(dataset, epochs=50)

错误消息是：

Epoch 1/50
Traceback (most recent call last):

  File "<ipython-input-129-f200a4503ff9>", line 1, in <module>
    runfile('D:/ai/collab-filter/dan.py', wdir='D:/ai/collab-filter')

  File "C:\Users\DELL\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 786, in runfile
    execfile(filename, namespace)

  File "C:\Users\DELL\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 110, in execfile
    exec(compile(f.read(), filename, 'exec'), namespace)

  File "D:/ai/myModel.py", line 256, in <module>
    model.fit(dataset, epochs=50)

  File "C:\Users\DELL\Anaconda3\lib\site-packages\tensorflow\python\keras\engine\training.py", line 694, in fit
    initial_epoch=initial_epoch)

  File "C:\Users\DELL\Anaconda3\lib\site-packages\tensorflow\python\keras\engine\training.py", line 1433, in fit_generator
    steps_name='steps_per_epoch')

  File "C:\Users\DELL\Anaconda3\lib\site-packages\tensorflow\python\keras\engine\training_generator.py", line 256, in model_iteration
    batch_size = int(nest.flatten(batch_data)[0].shape[0])

  File "C:\Users\DELL\Anaconda3\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 868, in __getitem__
    return self._dims[key]

IndexError: list index out of range

取消对示例的打印注释时（对我来说看起来不错）：

(<tf.Tensor: id=281899, shape=(), dtype=float32, numpy=0.0011093982>, <tf.Tensor: id=281900, shape=(), dtype=float32, numpy=0.008290171>, <tf.Tensor: id=281901, shape=(), dtype=float32, numpy=0.010696268>, <tf.Tensor: id=281902, shape=(), dtype=float32, numpy=-0.0068962937>, <tf.Tensor: id=281903, shape=(), dtype=float32, numpy=0.0020472356>, <tf.Tensor: id=281904, shape=(), dtype=float32, numpy=0.0041239075>, <tf.Tensor: id=281905, shape=(), dtype=float32, numpy=-0.0018036675>, <tf.Tensor: id=281906, shape=(), dtype=float32, numpy=-0.007521228>, <tf.Tensor: id=281907, shape=(), dtype=float32, numpy=0.012179799>, <tf.Tensor: id=281908, shape=(), dtype=float32, numpy=-0.008569455>, <tf.Tensor: id=281909, shape=(), dtype=float32, numpy=-0.005547243>, <tf.Tensor: id=281910, shape=(), dtype=float32, numpy=-0.024963537>, <tf.Tensor: id=281911, shape=(), dtype=float32, numpy=-0.0047834134>, <tf.Tensor: id=281912, shape=(), dtype=float32, numpy=-0.0073425>, <tf.Tensor: id=281913, shape=(), dtype=float32, numpy=-0.0049664816>, <tf.Tensor: id=281914, shape=(), dtype=float32, numpy=0.0012769673>, <tf.Tensor: id=281915, shape=(), dtype=float32, numpy=-0.008846987>, <tf.Tensor: id=281916, shape=(), dtype=float32, numpy=0.002845391>, <tf.Tensor: id=281917, shape=(), dtype=float32, numpy=-0.0012304187>, <tf.Tensor: id=281918, shape=(), dtype=float32, numpy=-0.0073605254>, <tf.Tensor: id=281919, shape=(), dtype=float32, numpy=-0.019149099>, <tf.Tensor: id=281920, shape=(), dtype=float32, numpy=0.0053162603>, <tf.Tensor: id=281921, shape=(), dtype=float32, numpy=0.00018294304>, <tf.Tensor: id=281922, shape=(), dtype=float32, numpy=-0.007135446>, <tf.Tensor: id=281923, shape=(), dtype=float32, numpy=0.019139009>, <tf.Tensor: id=281924, shape=(), dtype=float32, numpy=0.0031176396>, <tf.Tensor: id=281925, shape=(), dtype=float32, numpy=0.016997647>, <tf.Tensor: id=281926, shape=(), dtype=float32, numpy=-0.017783713>, <tf.Tensor: id=281927, shape=(), dtype=float32, numpy=-0.0033694915>, <tf.Tensor: id=281928, shape=(), dtype=float32, numpy=0.02030162>, <tf.Tensor: id=281929, shape=(), dtype=float32, numpy=-0.01870913>, <tf.Tensor: id=281930, shape=(), dtype=float32, numpy=-0.0057595233>, <tf.Tensor: id=281931, shape=(), dtype=float32, numpy=0.013816875>, <tf.Tensor: id=281932, shape=(), dtype=float32, numpy=-0.00463876>, <tf.Tensor: id=281933, shape=(), dtype=float32, numpy=-0.023181098>, <tf.Tensor: id=281934, shape=(), dtype=float32, numpy=0.0064159813>, <tf.Tensor: id=281935, shape=(), dtype=float32, numpy=-0.0018356718>, <tf.Tensor: id=281936, shape=(), dtype=float32, numpy=0.014198529>, <tf.Tensor: id=281937, shape=(), dtype=float32, numpy=-0.019970264>, <tf.Tensor: id=281938, shape=(), dtype=float32, numpy=-0.013106668>, <tf.Tensor: id=281939, shape=(), dtype=float32, numpy=0.01739781>, <tf.Tensor: id=281940, shape=(), dtype=float32, numpy=-0.0075084846>, <tf.Tensor: id=281941, shape=(), dtype=float32, numpy=-0.007515852>, <tf.Tensor: id=281942, shape=(), dtype=float32, numpy=0.008860749>, <tf.Tensor: id=281943, shape=(), dtype=float32, numpy=0.011078904>, <tf.Tensor: id=281944, shape=(), dtype=float32, numpy=0.0031385398>, <tf.Tensor: id=281945, shape=(), dtype=float32, numpy=0.00069636817>, <tf.Tensor: id=281946, shape=(), dtype=float32, numpy=0.016473386>, <tf.Tensor: id=281947, shape=(), dtype=float32, numpy=0.010464343>, <tf.Tensor: id=281948, shape=(), dtype=float32, numpy=0.009564337>, <tf.Tensor: id=281949, shape=(), dtype=float32, numpy=-0.00023193806>, <tf.Tensor: id=281950, shape=(), dtype=float32, numpy=-0.0043777116>, <tf.Tensor: id=281951, shape=(), dtype=float32, numpy=0.0033248402>, <tf.Tensor: id=281952, shape=(), dtype=float32, numpy=0.0020942744>, <tf.Tensor: id=281953, shape=(), dtype=float32, numpy=0.00989055>, <tf.Tensor: id=281954, shape=(), dtype=float32, numpy=0.000547247>, <tf.Tensor: id=281955, shape=(), dtype=float32, numpy=-0.0011691392>, <tf.Tensor: id=281956, shape=(), dtype=float32, numpy=-0.033643395>, <tf.Tensor: id=281957, shape=(), dtype=float32, numpy=-0.0014932752>, <tf.Tensor: id=281958, shape=(), dtype=float32, numpy=0.012660088>, <tf.Tensor: id=281959, shape=(), dtype=float32, numpy=0.0124913>, <tf.Tensor: id=281960, shape=(), dtype=float32, numpy=-0.010591994>, <tf.Tensor: id=281961, shape=(), dtype=float32, numpy=-0.030872872>, <tf.Tensor: id=281962, shape=(), dtype=float32, numpy=-0.0014752604>)



tf.Tensor(
[0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0], shape=(100,), dtype=int32)

Answer 1

我能够解决此问题。就像我之前说过的那样，我是TensorFlow的新手，我不知道tf.data.experimental.CsvDataset()的结果（我用它来读取我的数据）不能直接用于训练模型。结果采用64张量的元组形式，但是现在我知道训练示例必须至少具有2个维度（第一个可以为1），而第一个维度为batch_size，第二个维度为示例本身的大小。在我的代码中，第一个Tensor（例如<tf.Tensor: id=281899, shape=(), dtype=float32, numpy=0.0011093982>）被视为第一个训练示例，因此其形状为（），这导致了错误。当我尝试解决问题时，我做了一些更改，导致了其他错误，这使我找到了解决方案。每个示例都必须调整大小。这是我当前的功能版本，用于阅读培训示例（DATA_SIZE为64）：

def expMap(*item) :
   item2 = tf.reshape(item, [1, DATA_SIZE])
   return item2

def generate_trainExp(nameD):
   filename = os.path.join(DATA, "exps{}.txt".format(nameD))
   record_defaults = [tf.float32] * DATA_SIZE
   dataset = tf.data.experimental.CsvDataset(filename, record_defaults, field_delim=' ')
   dataset = dataset.map(expMap)
   return dataset

.....
data = generate_trainExp(filename)

在expMap()中，输入项也可以在重塑指令（item = tf.convert_to_tensor(item, dtype=tf.float32)）之前转换为64个浮点数的向量。

在Tensorflow Keras分类器中使用数据集时在model.fit（）方法中出现“ IndexError：列表索引超出范围”

1 个答案: