collaborative_SVD.py

# -*- coding: utf-8 -*-
"""
Created on Sat Nov 17 19:11:51 2018

@author: archit bansal
"""

import pandas as pd
import numpy as np
from scipy.sparse.linalg import svds
ratings_df=pd.read_csv('ratings.csv',sep=",")
movies_df=pd.read_csv('movies.csv',sep=",",encoding='latin-1')

A_df=ratings_df.pivot_table(index=['userId'],columns=['movieId'],values='rating',aggfunc=np.max)
A_df.fillna(0,inplace=True)
A=A_df.as_matrix()
user_rating_mean=np.mean(A,axis=1)
A_normalized=A-user_rating_mean.reshape(-1,1)

U,sigma,Vt=svds(A_normalized,k=50)
sigma=np.diag(sigma)
predicted_rating=np.dot(np.dot(U,sigma),Vt) + user_rating_mean.reshape(-1,1)
predicted_rating_df=pd.DataFrame(predicted_rating,columns=A_df.columns)

def recommend_movies(prediction_df,userID,movies_df,original_ratings_df,num_recommendations):
    user_row_number=userID-1
    sorted_user_prediction=predicted_rating_df.iloc[user_row_number].sort_values
    user_data=original_ratings_df[original_ratings_df.userId==(userID)]
    user_full= user_data.merge(movies_df,how='left',left_on='movieId',right_on='movieId')
    user_full=user_full.sort_values(['rating'],ascending=False)
                              
    print('user {} has already rated {} movies'.format(userID,user_full.shape[0]))
    print('recommendaing highest {0} predicted ratings movies not already rated'.format(num_recommendations))
    recommedations=(movies_df[~movies_df['movieId'].isin(user_full['movieId'])].merge
                              (pd.DataFrame(sorted_user_prediction).reset_index(),how='left',
                               left_on='movieId',right_on='movieId').rename(columns={user_row_number:'Predictions'})
                               .sort_values('Predictions',ascending=False).iloc[:num_recommendations,:-1])
    
    return (user_full,recommendations)
already_rated,predictions=recommend_movies(predicted_rating_df,2,movies_df,ratings_df,10)
already_rated=already_rated.head(10)
predictions=predictions