This is a very basic data generator to test recommender systems. A future version may simulate the actual sparseness of ratings data with a simple bootstrap function but for now, numpy generator does the job.

readme.md

RecData

To use this snippet, install faker:

pip install faker

Generate 100 users with 200 title features (default settings)

rec_data = RecData()
df = rec_data.generate_df()
df

Generate 20 users /w 12 title features

rec_data = RecData(n_users=20, n_features=12)
df = rec_data.generate_df()
df

Generate 500 users /w 3000 title features with a ratings scale of 0-5

rec_data = RecData(n_users=500, n_features=3000, ratings_scale=6)
df = rec_data.generate_df()
df

Share this with anyone you want 😉

rec_data.py

import numpy as np, pandas as pd
from faker import Factory

class RecData:
    
    ratings_scale = 2
    n_users       = 100
    n_features    = 200
    
    def __init__(self, **opts):        
        ## set defaults
        self.n_users = opts.get('n_users', 100)
        self.n_features = opts.get('n_features', 200)
        self.ratings_scale = opts.get('ratings_scale', 2) # 2 = binary 1 or 0
        
        ## dont forget to pip install faker
        self.faker = Factory.create()
        
    def generate_names(self):
        return [self.faker.name() for n in range(0, self.n_users)]
    
    def generate_fake_title(self):
        return " ".join([title.title() for title in self.faker.text().split()[:3]])
    
    def generate_titles(self):
        return [self.generate_fake_title() for n in range(0, self.n_features)]
    
    def generate_df(self):
        df = pd.DataFrame(
            np.random.randint(
                self.ratings_scale, 
                size  = (self.n_users, self.n_features)
            ), 
            columns = self.generate_titles(),
            index   = self.generate_names()
        )
        return df