Gromov-delta-estimation

A project provides various functions for calculating Gromov delta-hyperbolicity on a user-item datasets. Current implementation works with Amazon and MovieLens datasets. All calculus are performed with hypdelta module.

Repo structure

All variations of algos for calculating delta are located in lib/source/algo folder.
notebooks folder contains useful jupiter notebooks for visualizing statistics and charting.
scripts for running current configuration are in scripts folder.

Positional arguments

Builds csv file with experiment results 


csv_name:             Csv name
config_path:          Path to config file, where dependencies and default parameters are stored
min_batch:            Minimum batch size
min_rank:             Minimum rank
max_batch:            Maximum batch size
max_rank:             Maximum rank 
grid_batch_size:      Num measurements on batch
grid_rank:            Num measurements on rank
-v:                   Flag to print logs
-p:                   Flag to consider min_batch/max_batch as percents

Note that all essential directories (for datasets, csvs and svds) along with required dependencies should be set in the config file. See delta_config.yml for an example.

Example of usage

python scripts/est_delta.py -p -v experiments_delta.csv delta_config.yml 100 3000 100 3000 1 1