模型。fit() ValueError:检查模型目标时的错误:期望dense_21有形状(None, 1)，但有形状的数组(1708,66)

This the code I'm working on:

这是我正在编写的代码:

from __future__ import print_function
from keras.models import Sequential
from keras.layers import Dense
from sklearn.cross_validation import train_test_split
import numpy
numpy.random.seed(7)

data_pixels=np.genfromtxt("pixels_dataset.csv", delimiter=',')
classes_dataset=np.genfromtxt("labels.csv",dtype=np.str , delimiter='\t')
x_train, x_test, y_train, y_test = train_test_split(data_pixels, classes_dataset, test_size=0.3

x_train has a shape of (1708, 3072)

x_train有一个形状(1708,3072)

array([[ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  1.,  1.,  1.],
       [ 1.,  1.,  1., ...,  1.,  1.,  1.],
       ..., 
       [ 0.,  0.,  0., ...,  1.,  1.,  1.],
       [ 1.,  1.,  1., ...,  1.,  1.,  1.],
       [ 0.,  0.,  0., ...,  1.,  1.,  1.]])

y_train has a shape of (1708,)

y_train有一个形状(1708，)

array(['7', 'f', '3', ..., '6', 'o', 'O'], 
      dtype='|S5')

the characters of y_train are

y_train的字符是。

: , : ; ! è à ä Aa..Zz 0-9

::;!e一个Aa . .Zz 0 - 9

model = Sequential()
model.add(Dense(12, input_dim=3072, activation='relu'))
model.add(Dense(8, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])

l got error after executing the following :

在执行下列操作后，出现错误:

model.fit(x_train,y_train, epochs=150, batch_size=10)

the error is

错误的是

ValueError: could not convert string to float: A

l tried the following alternatives : 1)

我尝试了以下选项:1)

x_train=n.array(x_train)
y_train=n.array(y_train)

2)

2)

 model.fit(x_train,str(y_train), epochs=150, batch_size=10)

But l got the same error Then l tried another alternative

但是我有同样的错误，我尝试了另一种选择。

from sklearn.preprocessing import LabelBinarizer
encoder = LabelBinarizer()
y_train = encoder.fit_transform(y_train)

then l get a new error which is

然后我得到一个新的错误。

ValueError: Error when checking model target: expected dense_21 to have shape (None, 1) but got array with shape (1708, 66)

1 个解决方案

#1

Change the following lines of code:

更改以下代码行:

model.add(Dense(66, activation='softmax'))

and:

和:

model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

The problem lied in this that you wanted to predict a char which was coded as one-hot vector of length 66. In this case - you are setting your output to have desired length and you are using categorical_crossentropy loss and softmax activation.

问题在于，你想要预测一个字符，它被编码成一个长度为66的热矢量。在这种情况下，您将设置您的输出以获得所需的长度，并且您正在使用categorical_cross熵损失和softmax激活。