keras神经网络算法预测销量高低实践

基于样本数据周末,促销,天气,销量,进行深度学习。最终得出销量与周末、促销、天气的关系模型。并展示了神经网络算法对分类的预测。

以下为代码:

import pandas as pd

In [2]:

filename = ‘/Users/frontc/book/ppdam/sales_data.xls’

In [3]:

data = pd.read_excel(filename,index_col=u’序号’)

In [4]:

# 数据规整

data[data == u’好’] = 1

data[data == u’是’] = 1

data[data == u’高’] = 1

data[data !=1 ] = 0

In [5]:

x = data.iloc[:,:3].as_matrix().astype(int) # 自变量

y = data.iloc[:,3].as_matrix().astype(int) # 因变量

In [6]:

from keras.models import Sequential # 导入神经网络库

Using Theano backend.

In [7]:

from keras.layers.core import Dense,Activation # 导入层构建包

In [8]:

model = Sequential() # 初始化神经网络模型:线性叠加模型

In [9]:

model.add(Dense(input_dim=3,output_dim=10)) # 添加3个输入层节点,10个隐藏层节点

In [10]:

model.add(Activation(‘relu’)) # 设置激活函数为relu:Rectified Linear Units

In [11]:

model.add(Dense(output_dim=1)) # 设置输出层为1个节点

In [12]:

model.add(Activation(‘sigmoid’)) # 设置激活函数为sigmoid

In [14]:

model.compile(loss=’binary_crossentropy’,optimizer=’adam’) # 根据设置信息进行变异

In [15]:

model.fit(x,y,nb_epoch=1000,batch_size=10) # 迭代1000次,块大小10

Epoch 1/1000

34/34 [==============================] - 0s - loss: 0.6522

Epoch 2/1000

34/34 [==============================] - 0s - loss: 0.6502

…….

Epoch 999/1000

34/34 [==============================] - 0s - loss: 0.4472

Epoch 1000/1000

34/34 [==============================] - 0s - loss: 0.4474

Out[15]:

In [16]:

yp = model.predict_classes(x).reshape(len(y)) # predict_classes对x进行分类预测

32/34 [===========================>..] - ETA: 0s

In [19]:

#-\- coding: utf-8 --

def cm_plot(y, yp):

from sklearn.metrics import confusion_matrix

cm = confusion_matrix(y, yp)

import matplotlib.pyplot as plt

plt.matshow(cm, cmap=plt.cm.Greens)

plt.colorbar()

for x in range(len(cm)):

for y in range(len(cm)):

​ plt.annotate(cm[x,y], xy=(x, y), horizontalalignment=’center’, verticalalignment=’center’)

plt.ylabel(‘True label’)

plt.xlabel(‘Predicted label’)

return plt

In [20]:

cm_plot(y,yp).show() # 可以从混淆矩阵图中看出样本数为34个,其中预测正确的项目为26个

混淆矩阵