LensKit Evaluation Quickstart

This example project shows how to run and evaluate a custom LensKit recommender component using Gradle, the current recommended way to run LensKit evaluations. It is intended to serve as a template for you to use when you create new LensKit evaluations.

The key user files that you are likely to want to edit are:

• build.gradle: to configure the build, add dependencies, etc.
• eval.groovy: to change the lenskit evaluation that is run, perhaps by configuring different recommenders. = analyze-output.ipynb: to change the analysis of the output data in build, perhaps including the charts that are generated.

To run the evaluation, run:

./gradlew evaluate

As is typical with Gradle projects, all output files go in the build directory, where they can be removed with ./gradlew clean.

Viewing Analysis

We have provided a Jupyter notebook, analyze-output.ipynb, as an example of how to analyze the results of a LensKit evaluation in Python. To view this notebook, run jupyter notebook (or ipython notebook) after you've run the evaluation; it will automatically open a browser, and you can select the analyze-output notebook.

You can also run ./gradlew analyzeResults to perform a batch run of the analysis; the output will be in build/analysis.html.

The analysis requires the following software:

• Jupyter (formerly IPython) to view the notebook
• Pandas
• matplotlib

An easy way to get all of this software is to install the Anaconda Python distribution. It provides a complete scientific software stack for Python on Windows, OS X, and Linux.

The notebook is configured to run with the Python 2 kernel, but the code is entirely compatible with Python 3.

Example Scorer

There is a simple example scorer in src/main/java. This scorer includes a model that generates item mean ratings, and a scorer based on that model. You may find the model and predictor useful as starting points for your own predictors. The analysis script uses this scorer along with some well-known rating prediction algorithms.

More Information

More information on LensKit and its evaluator can be found on the LensKit web site.

Copyright

This project was created by the LensKit contributors, primarily Michael Ekstrand, Daniel Kluver, and John Riedl.

The files in this project may be freely modified, used, and distributed without restriction.

If further legal clarity is required, these files are licensed under Creative Commons CC0.