librec-auto is a Python tool for running recommender systems experiments.
It is built on top of the open-source LibRec package, and
can take advantage of the many algorithm and metric implementations there.
The basic element of librec-auto execution is the "study", which is a series
of experiments carried with a single algorithm, a single data set, and a set
of evaluation metrics. The experiments differ from each other by the hyperparameters
given to the algorithm. librec-auto allows such studies to be conducted with
minimal experimenter intervention, and supports such capabilities as:
- recommendation result re-ranking
- re-running evaluations without re-computing results
- fairness-aware metrics
- summarizing results in output graphs
- integration with Slack and Dropbox
More complete documentation is available at readthedocs:
The workflow of an study involves identifying appropriate data, creating training / test splits, implementing or choosing algorithms, running experiments (possibly with a range of different parameters), and reporting on the results.
Librec-auto uses an XML-based configuration system similar to Maven or Ant.
This directory contains the Python libraries for the librec_auto module. There are two other affiliated respositories:
- librec-auto-java: Contains the java source for the wrapper between LibRec and librec-auto, which is implemented in the
auto.jarfile. - librec-auto-demo2021: Contains sample data and configuration files that can be used to explore the functionality of librec-auto
/bin: Contains auxiliary scripts including the study setup wizard./librec_auto/jar: Contains the jar files for LibRec and the wrapper./librec_auto/rules: Contains the rules for translating configuration data to LibRec properties format./librec_auto/core: Contains the Python code for the project./librec_auto/docs: Contains documentation for the project/librec_auto/test: Contains the unit tests (not many right now)/librec_auto/library: Contains libraries of configuration information (look hear to see how to configure different algorithms)