概述
以房價預測爲例,使用numpy實現深度學習網絡--線性迴歸代碼。
數據鏈接:https://pan.baidu.com/s/1pY5gc3g8p-IK3AutjSUUMA
提取碼:l3oo
導入庫
import numpy as np
import matplotlib.pyplot as plt
加載數據
def LoadData():
#讀取數據
data = np.fromfile( './housing.data', sep=' ' )
#變換數據形狀
feature_names = ['CRIM', 'ZN', 'INDUS', 'CHAS', 'NOX', 'RM', 'AGE', 'DIS', 'RAD', 'TAX', 'PTRATIO', 'B', 'LSTAT', 'MEDV']
feature_num = len( feature_names )
data = data.reshape( [-1, feature_num] )
#計算數據最大值、最小值、平均值
data_max = data.max( axis=0 )
data_min = data.min( axis=0 )
data_avg = data.sum( axis=0 ) / data.shape[0]
#對數據進行歸一化處理
for i in range( feature_num ):
data[:, i] = ( data[:, i] - data_avg[i] ) / ( data_max[i] - data_min[i] )
#劃分訓練集和測試集
ratio = 0.8
offset = int( data.shape[0] * ratio )
train_data = data[ :offset ]
data_test = data[ offset: ]
return data_train, data_test
模型設計
class Network( object ):
'''
線性迴歸神經網絡類
'''
def __init__( self, num_weights ):
'''
初始化權重和偏置
'''
self.w = np.random.randn( num_weights, 1 ) #隨機初始化權重
self.b = 0.
def Forward( self, x ):
'''
前向訓練:計算預測值
'''
y_predict = np.dot( x, self.w ) + self.b #根據公式,計算預測值
return y_predict
def Loss( self, y_predict, y_real ):
'''
計算損失值:均方誤差法
'''
error = y_predict - y_real #誤差
cost = np.square( error ) #代價函數:誤差求平方
cost = np.mean( cost ) #求代價函數的均值(即:MSE法求損失)
return cost
def Gradient( self, x, y_real ):
'''
根據公式,計算權重和偏置的梯度
'''
y_predict = self.Forward( x ) #計算預測值
gradient_w = ( y_predict - y_real ) * x #根據公式,計算權重的梯度
gradient_w = np.mean( gradient_w, axis=0 ) #計算每一列的權重的平均值
gradient_w = gradient_w[:, np.newaxis] #reshape
gradient_b = ( y_predict - y_real ) #根據公式,計算偏置的梯度
gradient_b = np.mean( gradient_b ) #計算偏置梯度的平均值
return gradient_w, gradient_b
def Update( self, gradient_w, gradient_b, learning_rate=0.01 ):
'''
梯度下降法:更新權重和偏置
'''
self.w = self.w - gradient_w * learning_rate #根據公式,更新權重
self.b = self.b - gradient_b * learning_rate #根據公式,更新偏置
def Train( self, x, y, num_iter=100, learning_rate=0.01 ):
'''
使用梯度下降法,訓練模型
'''
losses = []
for i in range( num_iter ): #迭代計算更新權重、偏置
#計算預測值
y_predict = self.Forward( x )
#計算損失
loss = self.Loss( y_predict, y )
#計算梯度
gradient_w, gradient_b = self.Gradient( x, y )
#根據梯度,更新權重和偏置
self.Update( gradient_w, gradient_b, learning_rate )
#打印模型當前狀態
losses.append( loss )
if ( i+1 ) % 10 == 0:
print( 'iter = {}, loss = {}'.format( i+1, loss ) )
return losses
模型訓練
#獲取數據
train_data, test_data = LoadData()
x_data = train_data[:, :-1]
y_data = train_data[:, -1:]
#創建網絡
net = Network( 13 )
num_interator = 1000
learning_rate = 0.01
#進行訓練
losses = net.Train( x_data, y_data, num_interator, learning_rate )
#畫出損失函數變化趨勢
plot_x = np.arange( num_interator )
plot_y = losses
plt.plot( plot_x, plot_y )
plt.show()