replay-rec 0.10.0

RecSys Library

RePlay

RePlay is a library providing tools for all stages of creating a recommendation system, from data preprocessing to model evaluation and comparison.

RePlay uses PySpark to handle big data.

You can

• Filter and split data
• Train models
• Optimize hyper parameters
• Evaluate predictions with metrics
• Combine predictions from different models
• Create a two-level model

Documentation is available here.

Table of Contents

• Installation
• Quickstart
• Resources
• Contributing to RePlay

Installation

Use Linux machine with Python 3.7-3.9, Java 8+ and C++ compiler.

pip install replay-rec

To get the latest development version or RePlay, install it from the GitHab repository. It is preferable to use a virtual environment for your installation.

If you encounter an error during RePlay installation, check the troubleshooting guide.

Quickstart

from rs_datasets import MovieLens

from replay.data_preparator import DataPreparator, Indexer
from replay.metrics import HitRate, NDCG
from replay.models import ItemKNN
from replay.session_handler import State
from replay.splitters import UserSplitter

spark = State().session

ml_1m = MovieLens("1m")

# data preprocessing
preparator = DataPreparator()
log = preparator.transform(
    columns_mapping={
        'user_id': 'user_id',
        'item_id': 'item_id',
        'relevance': 'rating',
        'timestamp': 'timestamp'
    }, 
    data=ml_1m.ratings
)
indexer = Indexer(user_col='user_id', item_col='item_id')
indexer.fit(users=log.select('user_id'), items=log.select('item_id'))
log_replay = indexer.transform(df=log)

# data splitting
user_splitter = UserSplitter(
    item_test_size=10,
    user_test_size=500,
    drop_cold_items=True,
    drop_cold_users=True,
    shuffle=True,
    seed=42,
)
train, test = user_splitter.split(log_replay)

# model training
model = ItemKNN()
model.fit(train)

# model inference
recs = model.predict(
    log=train,
    k=K,
    users=test.select('user_idx').distinct(),
    filter_seen_items=True,
)

# model evaluation
metrics = Experiment(test,  {NDCG(): K, HitRate(): K})
metrics.add_result("knn", recs)

Resources

Usage examples

1. 01_replay_basics.ipynb - get started with RePlay.
2. 02_models_comparison.ipynb - reproducible models comparison on MovieLens-1M dataset.
3. 03_features_preprocessing_and_lightFM.ipynb - LightFM example with pyspark for feature preprocessing.
4. 04_splitters.ipynb - An example of using RePlay data splitters.
5. 05_feature_generators.ipynb - Feature generation with RePlay.

Videos and papers

• Video guides:

  • Replay for offline recommendations, AI Journey 2021

• Research papers:

  • Yan-Martin Tamm, Rinchin Damdinov, Alexey Vasilev Quality Metrics in Recommender Systems: Do We Calculate Metrics Consistently?

Contributing to RePlay

For more details visit development section in docs