Keras 和输入层

Question

所以我正在尝试使用 Keras 学习 ANN，因为我听说 Theano 或 TensorFlow 更简单。我有很多问题，第一个是关于输入层的。

到目前为止，我有这行代码作为输入：

model.add(Dense(3 ,input_shape=(2,), batch_size=50 ,activation='relu'))

现在我要添加到模型中的数据具有以下形状：

    Index(['stock_price', 'stock_volume', 'sentiment'], dtype='object')
[[  3.01440000e+02   7.87830000e+04   0.00000000e+00]
 [  3.01440000e+02   7.87830000e+04   0.00000000e+00]
 [  3.01440000e+02   7.87830000e+04   1.42857143e-01]
 [  3.01440000e+02   7.87830000e+04   5.88235294e-02]
 [  3.01440000e+02   7.87830000e+04   0.00000000e+00]
 [  3.01440000e+02   7.87830000e+04   0.00000000e+00]
 [  3.01440000e+02   7.87830000e+04   0.00000000e+00]
 [  3.01440000e+02   7.87830000e+04   0.00000000e+00]
 [  3.01440000e+02   7.87830000e+04   0.00000000e+00]
 [  3.01440000e+02   7.87830000e+04   5.26315789e-02]]

我想建立一个模型，看看我能否找到股票价格和推文情绪之间的相关性，我只是在其中投入了交易量，因为最终，我想看看它是否也能找到与之相关的模式。

所以我的第二个问题是在使用几个不同的参数运行我的输入层之后，我遇到了这个我无法解释的问题。所以当我运行这一行时：

model.add(Dense(3 ,input_shape=(2,), batch_size=50 ,activation='relu'))

使用以下行我得到这个输出错误：

ValueError: Error when checking model input: expected dense_1_input to have shape (50, 2) but got array with shape (50, 3)

但是当我将输入形状更改为请求的“3”时，我收到此错误：

ValueError: Error when checking model target: expected dense_2 to have shape (50, 1) but got array with shape (50, 302)

为什么错误信息上的2变成了'302'？

我可能忽略了一些非常基本的问题，因为这是我尝试实现的第一个神经网络，因为我之前只使用过 Weka 的应用程序。

这里是我完整代码的副本：

from keras.models import Sequential, Model
from keras.layers import Dense, Activation, Input
from keras.optimizers import SGD
from keras.utils import np_utils
import pymysql as mysql
import pandas as pd
import config

import numpy
import pprint

model = Sequential()
try:
    sql = "SELECT stock_price, stock_volume, sentiment FROM tweets LIMIT 50"
    con = mysql.connect(config.dbhost, config.dbuser, config.dbpassword, config.dbname, charset='utf8mb4', autocommit=True)
    results = pd.read_sql(sql=sql, con=con, columns=['stock_price', 'stock_volume', 'sentiment'])
finally:
    con.close()

npResults = results.as_matrix()
cols = np_utils.to_categorical(results['stock_price'].values)
data = results.values

print(cols)
# inputs:
# 1st = stock price
# 2nd = tweet sentiment
# 3rd = volume
model.add(Dense(3 ,input_shape=(3,), batch_size=50 ,activation='relu'))
model.add(Dense(20, activation='linear'))
sgd = SGD(lr=0.3, decay=0.01, momentum=0.2)

model.compile(loss='sparse_categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])

model.summary()
model.fit(x=data, y=cols, epochs=100, batch_size=100, verbose=2)

编辑：

这是我从控制台得到的所有输出：

    C:\Users\Def\Anaconda3\python.exe C:/Users/Def/Dropbox/Dissertation/ann.py
Using Theano backend.
C:\Users\Def\Dropbox\Dissertation
[[ 0.  0.  0. ...,  0.  0.  1.]
 [ 0.  0.  0. ...,  0.  0.  1.]
 [ 0.  0.  0. ...,  0.  0.  1.]
 ..., 
 [ 0.  0.  0. ...,  0.  0.  1.]
 [ 0.  0.  0. ...,  0.  0.  1.]
 [ 0.  0.  0. ...,  0.  0.  1.]]
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_1 (Dense)              (50, 3)                   12        
_________________________________________________________________
dense_2 (Dense)              (50, 20)                  80        
=================================================================
Traceback (most recent call last):
  File "C:/Users/Def/Dropbox/Dissertation/ann.py", line 38, in <module>
    model.fit(x=data, y=cols, epochs=100, batch_size=100, verbose=2)
  File "C:\Users\Def\Anaconda3\lib\site-packages\keras\models.py", line 845, in fit
    initial_epoch=initial_epoch)
  File "C:\Users\Def\Anaconda3\lib\site-packages\keras\engine\training.py", line 1405, in fit
    batch_size=batch_size)
  File "C:\Users\Def\Anaconda3\lib\site-packages\keras\engine\training.py", line 1299, in _standardize_user_data
    exception_prefix='model target')
  File "C:\Users\Def\Anaconda3\lib\site-packages\keras\engine\training.py", line 133, in _standardize_input_data
    str(array.shape))
ValueError: Error when checking model target: expected dense_2 to have shape (50, 20) but got array with shape (50, 302)
Total params: 92.0
Trainable params: 92
Non-trainable params: 0.0
_________________________________________________________________

Process finished with exit code 1

Answer 1

我认为您使用了错误的指标：sparse_categorical_crossentropy 你有什么理由喜欢这个而不是正常的：categorical_crossentropy？

使用categorical_crossentropy 时，您应该以1-hot 编码方式对目标进行编码（例如：cols = np_utils.to_categorical(results['stock_price'].values)）。

另一方面，sparse_categorical_crossentropy 使用基于整数的标签。

所以要么使用：

cols = np_utils.to_categorical(results['stock_price'].values)

与

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

和一个（num-categories）神经元的输出层

或使用：

cols = results['stock_price'].values.astype(np.int32)

与

model.compile(loss='sparse_categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])

和一个单神经元输出层。