This module holds code for my master's thesis. The goal is to create a recommender system that not only learns user item preferences but also takes item inventory into account.
This is intended to be used as a sandbox for testing different recommendation systems and evaluating their performance by simulating users buying items from inventory.
The important classes are:
- UIRData - used to store and preprocess data
- Inventory - holds inventory data
- BaseRecommender - base class for recommenders
- MatrixFactorization - implementation of matrix factorization recommender system
- BaseSimulator - base class for simulations
Other interesting functionalities are running repeated simulations on different random initializations, training recommender systems and simulating them in parallel (works only on Unix systems), diverse plotting functions, more recommender systems and simulations.
Use Python 3.5 or higher.
Clone the repository and pip install the module.
git clone https://github.com/robertcv/pyrec
cd pyrec
pip install .If you would like to change existing code use pip install -e ..
If you are on Windows and have trouble installing any of the dependencies
I would recommend using Anaconda.
A simple use case would be:
from pyrec.data import UIRData
from pyrec.inventory import Inventory
from pyrec.recs.mf import MatrixFactorization
from pyrec.sims.rand import BestSimulator
# load data from csv, preprocess, split into train, validation and test
RATINGS_FILE = "data/MovieLens/ml-latest-small/ratings.csv"
uir_data = UIRData.from_csv(RATINGS_FILE)
# initialize matrix factorization, train the model
mf = MatrixFactorization()
mf.fit(uir_data)
# create inventory from rating frequency if no explicit inventory data is given
inv = Inventory(uir_data)
# initialize and run the simulation
sim = BestSimulator("rand", uir_data, mf, inv)
results = sim.run()
print(results)For more examples see the examples folder.