python-3.x - Model.fit 中自定义生成器的 Tensorflow Keras 问题

Question

我创建了一个自定义生成器tensorflow.keras.utils.Sequence并尝试拟合一个简单的模型。

import tensorflow as tf
import keras
import numpy as np
from tensorflow.keras.utils import Sequence
from keras import Sequential
from keras.layers import InputLayer, Dense

class MyDataGenerator(Sequence):
  def __init__(self, df, x_col='filename', y_col='class',
                 batch_size=32, path='./', num_classes=None, shuffle=True,
                 dim=(634,513,1), nfft=1024, hstep=256, sr=16000):
    self.batch_size = batch_size
    self.df = df
    self.indices = self.df.index.tolist()
    self.num_classes = num_classes
    self.path = path
    self.shuffle = shuffle
    self.x_col = x_col
    self.y_col = y_col
    self.on_epoch_end()
    self.dim = dim
    self.nfft = nfft
    self.hstep = hstep
    self.sr = sr

  def __getitem__(self, index):
    index = self.index[index * self.batch_size:(index + 1) * self.batch_size]
    batch = [self.indices[k] for k in index]
    X = np.zeros((self.batch_size, *self.dim), dtype=np.float32)
    y = np.zeros((self.batch_size,), dtype=np.uint32)
    return X, y
    
  def __len__(self):
    return math.ceil(len(self.indices) / self.batch_size)

  def on_epoch_end(self):
    self.index = np.arange(len(self.indices))
    if self.shuffle == True:
      np.random.shuffle(self.index)

train_datagen = MyDataGenerator(df_training)
valid_datagen = MyDataGenerator(df_validation, shuffle=False)

x,y = train_datagen[27]
print(x.shape, y.shape, x.dtype, y.dtype, type(x), type(y))
print(y)

1. (32, 634, 513, 1) (32,) float32 uint32 <class 'numpy.ndarray'> <class 'numpy.ndarray'>
2. [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]

model = Sequential(name='Test_model')
model.add(InputLayer((634, 513, 1), name='Input'))
model.add(Flatten(name='Flatten'))
model.add(Dense(1, activation='sigmoid', name='Output'))
model.summary()
model.compile(optimizer='adam', loss='binary_crossentropy',
              metrics=['accuracy'])
model.fit(
    train_datagen,
    epochs=1,
    verbose=0,
    validation_data=valid_datagen
    )

我得到了ValueError

Model: "Test_model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
Flatten (Flatten)            (None, 325242)            0         
_________________________________________________________________
Output (Dense)               (None, 1)                 325243    
=================================================================
Total params: 325,243
Trainable params: 325,243
Non-trainable params: 0
_________________________________________________________________
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-203-533c619e3ae0> in <module>()
     11     epochs=1,
     12     verbose=0,
---> 13     validation_data=valid_datagen
     14     )

3 frames
/usr/local/lib/python3.7/dist-packages/keras/engine/data_adapter.py in select_data_adapter(x, y)
    976         "Failed to find data adapter that can handle "
    977         "input: {}, {}".format(
--> 978             _type_name(x), _type_name(y)))
    979   elif len(adapter_cls) > 1:
    980     raise RuntimeError(

ValueError: Failed to find data adapter that can handle input: <class '__main__.MyDataGenerator'>, <class 'NoneType'>

无法弄清楚为什么模型不想使用生成器进行训练。数据类型正确。输入和输出也是。

如果尝试在没有生成器的情况下执行此操作，则模型可以正常工作，但由于数据集庞大，无法删除生成器。

X = np.zeros((32, 634, 513, 1), dtype=np.float32)
y = np.zeros((32,), dtype=np.uint32)
model = Sequential(name='Test_model')
model.add(InputLayer((634, 513, 1), name='Input'))
model.add(Flatten(name='Flatten'))
model.add(Dense(1, activation='sigmoid', name='Output'))
model.summary()
model.compile(optimizer='adam', loss='binary_crossentropy',
              metrics=['accuracy'])
model.fit(
    X,y,
    epochs=1,
    verbose=1,
    #validation_data=valid_datagen
    )

Model: "Test_model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
Flatten (Flatten)            (None, 325242)            0         
_________________________________________________________________
Output (Dense)               (None, 1)                 325243    
=================================================================
Total params: 325,243
Trainable params: 325,243
Non-trainable params: 0
_________________________________________________________________
1/1 [==============================] - 14s 14s/step - loss: 0.6931 - accuracy: 1.0000
<keras.callbacks.History at 0x7f74adf34510>

张量流版本：2.5.0

Keras 版本：2.4.3

标签： python-3.xtensorflowkeras