基于模型融合的推荐系统实现(3):模型融合

基本思路很简单，最小二乘法就好了:

我们假设两个算法得到的结果权重分别是a,b利用最小二乘法和我们分出来的第二部分数据就可以获取a,b使得误差最小。其实最小二乘法就是求一个广义的逆即可。最后的RMSE比起单一的模型有所提高,变成了(0.86~~~~)

import numpy as npfrom code import read_filefrom SVD import construct_matrix,get_svd_predict,svd_predict,get_meanfrom PCA import get_train,get_pq,predictdef merge_models(index,mean=0):    train =  get_train(path=r'smaller_test.txt')#x    p,q,bu,bi = get_pq(index=index)    svd_predictions = get_svd_predict(index)    A = np.zeros((len(train),2))    Y = np.zeros((len(train)))    for user_id,iid,r in read_file(r'smaller_test.txt'):        pr_lfm = predict(user_id,iid,p,q,bu,bi)         pr_svd = svd_predict(user_id,iid,svd_predictions)        pr_lfm = pr_lfm if pr_lfm else 3        A[user_id-1] = np.array([pr_lfm,pr_svd+mean])        Y[user_id-1] = np.array([r])    alpha_mat = np.dot(np.dot(np.linalg.inv(np.dot(np.transpose(A), A)), np.transpose(A)), Y)    return alpha_matdef REMS(index,mean=0):    svd_predictions = get_svd_predict(index)    p,q,bu,bi = get_pq(index=index)    error = 0    cnt = 0    alpha1,alpha2 = merge_models(index,mean=mean)    cutoff = {1:-0.8,2:-1,3:0,4:0.3,5:0}    for uid,iid in read_file(r'test.txt',num = 2):        pr_lfm = predict(uid,iid,p,q,bu,bi)        pr_svd = svd_predict(uid,iid,svd_predictions)        if pr_lfm == None or pr_svd==None:continue        pr = alpha1*pr_lfm+alpha2*(pr_svd+mean)        if pr<1:pr = 1.0        if pr>5:pr = 5.0        cnt += 1        yield prdef write_ans(w_path,data):    with open(w_path,'w'):        pass    with open(w_path,'a') as file:            for r in data:                file.write('{0:.3f}\n'.format(r))if __name__ == '__main__':    #merge_models()    train = get_train(path=r'smaller_test.txt')    a,b,mean = get_mean(train)    write_ans('scores-5.txt',REMS(5,mean))    write_ans('scores-20.txt',REMS(20,mean))    write_ans('scores-50.txt',REMS(50,mean))