ransac 算法python源码实现

ransac算法最早是出现统计学中的，后面逐渐应用到image 领域。其对噪声有很好的鲁棒性。
算法如下：

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
from sklearn.linear_model import LinearRegression
from sklearn.preprocessing import PolynomialFeatures
import numpy as np
trans=PolynomialFeatures(degree=2)
linear_clf=LinearRegression()
def ransac(data,model,n,k,t,d):
    '''
    data:二维数据
    model:训练好的模型
    n:随机抽取样本数目
    k:迭代次数
    t:阈值
    d:测试集大于多少才认为其是好的模型
    '''
    data_size=len(data)
    epoch=0
    best_model=None   
    besterr=np.inf
    best_inlier_idxs=None
    while epoch<k:
        maybe_idxs,test_idxs=shuffle_data(data_size, n)
        maybe_inliers=data[maybe_idxs,:]
        test_inlier=data[test_idxs,:]
        maybeModel=linear_clf.fit(trans.fit_transform(maybe_inliers[:,:-1]),maybe_inliers[:,-1])
        test_error,_=get_error(maybeModel, test_inlier[:,:-1], test_inlier[:,-1])
        also_idxs=test_idxs[test_error<t]
        also_inliers=data[also_idxs,:]
        if len(also_inliers)>d:
            better_data=np.concatenate((maybe_inliers,also_inliers))
            betterModel=linear_clf.fit(trans.fit_transform(better_data[:,:-1]),better_data[:,-1])
            _,thisError=get_error(betterModel,better_data[:,:-1],better_data[:,-1])
            if thisError<besterr:
                best_model=betterModel
                besterr=thisError
                best_inlier_idxs=np.concatenate((maybe_idxs,also_idxs))
        epoch+=1
    if best_model is None:
        raise ValueError("无法拟合出model")
    else:
    	return best_model,besterr,best_inlier_idxs
def shuffle_data(data_row,n):
    idxs=np.arange(data_row)
    np.random.shuffle(idxs)
    return idxs[:n],idxs[n:]
def get_error(model,test,y_true):
    y_predict=model.predict(test)
    error=np.sqrt((y_predict-y_true)**2)
    mean_error=np.mean(error)
    return error,mean_error