逻辑回归的分布式实现 [Logistic Regression / Machine Learning / Spark ]

1- 问题提出

 

---

 

2- 逻辑回归

 

---

 

3- 理论推导

 

---

 

4- Python/Spark实现

# -*- coding: utf-8 -*-
from pyspark import SparkContext
from math import *

theta = [0, 0, 0]    #初始theta值
alpha = 0.001    #学习速率

def inner(x, y):
    return sum([i*j for i,j in zip(x,y)])
        
def func(lst):
    h = (1 + exp(-inner(lst, theta)))**(-1)
    return map(lambda x: (h - lst[-1]) * x, lst[:-1])


sc = SparkContext('local')

rdd = sc.textFile('/home/freyr/logisticRegression.txt')\
        .map(lambda line: map(float, line.strip().split(',')))\
        .map(lambda lst: [1]+lst)


for i in range(400):
    partheta = rdd.map(func)\
                   .reduce(lambda x,y: [i+j for i,j in zip(x,y)])

    for j in range(3):
        theta[j] = theta[j] - alpha * partheta[j]

print 'theta = %s' % theta

 PS: logisticRegression.txt