Keras和输入层

Question

所以我试图学习与凯拉斯的ANN，因为我听说Theano或TensorFlow更简单。我有许多问题是第一个与输入层有关的问题。

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')) 

现在我想添加到模型中的数据如下形状：

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]] 

我要打一个模型，看看我能否找到股票价格和tweet情绪之间的相关性，我只是把数量放在那里，因为最终，我想看看它是否也能找到一个模式。

所以，我的第二个问题是运行我的输入层与几个不同的参数后，我得到这个问题，我不能解释。所以，当我跑这条线：

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

与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，当我输入形状改变成请求“3”我得到这个错误：

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

为什么在错误信息中2变成'302'？

我可能忽略了一些真正的基本问题，因为这是我尝试实现的第一个神经网络，因为我以前只使用Weka的应用程序。

反正这里是我的全部代码的副本：

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：

这里是所有的输出我得到的FOM控制台：

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

我认为你正在使用错误的度量标准：sparse_categorical_crossentropy 是否有这样的理由比较正常：categorical_crossentropy？

当使用categorical_crossentropy时，应该使用单热编码方式（例如使用cols = np_utils.to_categorical(results['stock_price'].values)）编码目标。

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

因此，无论使用：

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

与

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

和（NUM-类别）神经元

或使用一个输出层：

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

与

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

和单神经元输出层。