machine_learn/linear/gradient_linear.py

#!/usr/bin/python
# -*- coding: UTF-8 -*-
'''
最简单的mse
'''
import sys
reload(sys)
sys.setdefaultencoding('utf-8')

import random
import numpy as np
from sklearn.linear_model import LinearRegression
from sklearn import metrics
from draw import draw_util


def read_data(path):
    with open(path) as f :
        lines=f.readlines()
    lines=[eval(line.strip()) for line in lines]
    return lines


def cal_step_pow(data, w, b=3):
    p = [(w*item[0][0] + b - item[1][0])*item[0][0]*2  for item in data]
    sum_p = sum(p)
    return sum_p/len(data)


def cal_step_pow_b(data, w, b=1):
    p = [(w*item[0][0] + b - item[1][0])*2  for item in data]
    sum_p = sum(p)
    return sum_p/len(data)


def cal_mse(data, w, b=1):
    sum_p = sum([(w * item[0][0] + b - item[1][0]) * (w * item[0][0] + b - item[1][0]) for item in data])
    return sum_p / len(data)


def train():
    train_data = read_data('train_data')
    w = random.uniform(-50, 50)
    for i in range(50):
        step = cal_step_pow(train_data, w)*0.01
        mse = cal_mse(train_data, w)
        print w, step, mse

        w = w - step
    return w


def train_b(w):
    train_data = read_data('train_data')
    b = random.uniform(-50, 50)
    for i in range(1000):
        step = cal_step_pow_b(train_data, w, b)*0.01
        mse = cal_mse(train_data, w, b)
        print b, step, mse

        b = b - step
    return b


if __name__ == '__main__':
    w = train()
    print "__________"
    b = train_b(w)
    print "__________"
    print w,b

    train_data = read_data('train_data')
    X, y = zip(*train_data)
    X = np.array(X)
    y = np.array(y)
    model = LinearRegression()
    # 一调用这个函数，就会不停地找合适的w和b 直到误差最小
    model.fit(X, y)
    print model.coef_, model.intercept_