欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页

Keras 报错When using data tensors as input to a model, you should specify the steps_per_epoch argument

程序员文章站 2022-05-26 19:00:33
...

报错解决:valueError: When using data tensors as input to a model, you should specify the steps_per_epoch argument.


Keras小白开始入手深度学习的时候,使用Sequence()建模的很舒服,突然有一天要使用到Model()的时候,就开始各种报错。

from keras.models import Sequential
from keras.layers import Dense, Activation

...

model = Sequential([
    Dense(32, input_shape=(784,)),
    Dense(10),
    Activation('softmax'),
])
history = model.fit(data_x_train, data_y_train, epoch=epoch, 
					batch_size=batch_size, 
					validation_data=(data_x_val, data_y_val),
					callbacks=callbacks, verbose=1, shuffle=True)

等到了使用Model()来建模的时候,需要指定input:

from keras.models import Sequential, Model
from keras.layers import Dense, Activation, Input

...

input1 = Input(batch_shape=(None, 28, 28), dtype=tf.float32)
des1 = Dense(32)(input1)
des2 = Dense(10)(des1)
y = Activation('softmax')(des2)

model = Model(inputs=input1, outputs=y)
history = model.fit(data_x_train, data_y_train, epoch=epoch, 
					batch_size=batch_size, 
					validation_data=(data_x_val, data_y_val),
					callbacks=callbacks, verbose=1, shuffle=True)

到开始fit的时候就开始报错

ValueError: When using data tensors as input to a model, you should specify the steps_per_epoch argument.

真的是找遍各种百度都解决不了的问题,我看Keras的教程中也有使用Model()建模然后使用batch_size训练的,百思不得其解。最后用Sequence蒙混过关,直到过了半年又得改成Model,我才发现:
Keras 报错When using data tensors as input to a model, you should specify the steps_per_epoch argument
怪不得说是tensor as input,我这里用了tf.float32,那可不就是变成tensor data了嘛,解决方法:

import numpy as np
from keras.models import Sequential, Model
from keras.layers import Dense, Activation, Input

...

input1 = Input(batch_shape=(None, 28, 28), dtype=np.float32)
des1 = Dense(32)(input1)
des2 = Dense(10)(des1)
y = Activation('softmax')(des2)

把tf改成np就可以解决问题

需要注意的是,只要在代码中出现了数据的reshape用到了tf,那么数据就变成了tensor data格式,例如

#data_x_train.shape = (256, 28)
tf.reshape(data_x_train, (256, 28, 1), name=tf.float32) 
tf.reshape(data_x_train, (256, 28, 1), name=np.float32) 

上述的两种写法都会使得数据从array或dataFrame或list变成tensor,所以为了避免使用steps_per_epoch(好多时候有问题都不知道怎么改,反正我是不喜欢用这个),可以用numpy的reshape方法来修改数据的维度:

# type(data_x_train) = DataFrame
data_x_train = np.array(data_x_train).reshape(256, 28, 1) 

之前有看到博主说删掉验证集可以解决这个问题,但我那时候也没成功,可能就是因为reshape的时候用了tf.float32

裂开




                  ——2021.2.25 随笔