Recommendation using ALS for implicit data. Code for Medium Blog

ALS_implementation.py

# -*- coding: utf-8 -*-
"""
Created on Sun Jun 23 22:20:58 2019

@author: himansh
"""
#import libraries
import sys
import pandas as pd
import numpy as np
import scipy.sparse as sparse
from scipy.sparse.linalg import spsolve
import random



from sklearn.preprocessing import MinMaxScaler

import implicit 
from datetime import datetime, timedelta


#Data Preprocessing
def create_data(datapath,start_date,end_date):
    df=pd.read_csv(datapath)
    df=df.assign(date=pd.Series(datetime.fromtimestamp(a/1000).date() for a in df.timestamp))
    df=df.sort_values(by='date').reset_index(drop=True) # for some reasons RetailRocket did NOT sort data by date
    df=df[(df.date>=datetime.strptime(start_date,'%Y-%m-%d').date())&(df.date<=datetime.strptime(end_date,'%Y-%m-%d').date())]
    df=df[['visitorid','itemid','event']]
    return df


#Download the kaggle RetailRocket data and give the events.csv file path
datapath= 'events.csv'
data=create_data(datapath,'2015-5-3','2015-5-18')
data['visitorid'] = data['visitorid'].astype("category")
data['itemid'] = data['itemid'].astype("category")
data['visitor_id'] = data['visitorid'].cat.codes
data['item_id'] = data['itemid'].cat.codes

data['event']=data['event'].astype('category')
data['event']=data['event'].cat.codes



sparse_item_user = sparse.csr_matrix((data['event'].astype(float), (data['item_id'], data['visitor_id'])))
sparse_user_item = sparse.csr_matrix((data['event'].astype(float), (data['visitor_id'], data['item_id'])))

#Building the model
model = implicit.als.AlternatingLeastSquares(factors=20, regularization=0.1, iterations=20)


alpha_val = 40
data_conf = (sparse_item_user * alpha_val).astype('double')


model.fit(data_conf)

###USING THE MODEL

#Get Recommendations
user_id =   14
recommended = model.recommend(user_id, sparse_user_item)
print(recommended)




#Get similar items
item_id = 7
n_similar = 3
similar = model.similar_items(item_id, n_similar)
print(similar)

How shall we tune the parameters? e.g. why did you chose these parameters? factors=20, regularization=0.1, iterations=20

There are always default values or those that usually work "better" and generally these values can be chosen at random (in the case of some parameters because others penalize the algorithm and "hinder" the recommendations). In case you have a metric that allows you to measure the performance of your model, I would advise you to use something like True Positive or the calculation of some distance, you can also calculate the average distance

in the paper ( collaborative filtering implicit dataset ), they multiply the r_ui by alpha to find the confidence value

Then shouldn't data_conf be defined as "1+ alpha_val X sparse_etc" ? instead of only alpha_val X sparse_... ?