/GSD-Estimator

Primary LanguagePython

GSD-Estimator

Estimator for Generalized Score Distibution's Distibution's parameters using OpenCL for accelerating calculation on GPI.

Requirements

pip install -r requirements.txt

Usage

  • Script takes scores.csv file and gsd_prob_grid in pandas or numpy format as input
  • output file will be a csv file with results decribed by values
    • idx - index of sample
    • psi - estimated mean (rating)
    • rho - estimated variance
    • log_likelihood - logarithm of likelihood value
  • While main.py is implementation with timers and debug prints, the same can be achieved with one simple function gsdest.start() inside another code that takes only samples and grid files as input and handles everything else: 
    import gsdest

gsdest.start()

TODO

  1. Optimize finding max likelihood loop to be less bad DONE
  2. Manage memory to not allocate too much in VRAM
  3. Manage input values not to use default files DONE
  4. Refactor main to functions DONE
  5. Make normal logger instead of simple prints DONE