polixir / d3pe

D3PE (Deep Data-Driven Policy Evaluation) aims to evaluation a large set of candidate policies from a fixed dataset to select best ones.

Introduction

D3PE (Deep Data-Driven Policy Evaluation) aims to evaluation a large set of candidate policies by a fix dataset to select best ones.

Supported Algorithms

• FQE
• MBOPE
• IS (WIS)
• Doubly-Robust

Installation

pip install -e .

or

pip install git+https://agit.ai/Polixir/d3pe.git

Usage

from d3pe.utils.data import OPEDataset
from d3pe.utils.func import get_evaluator_by_name

policy = ... # load your policy, API in d3pe.evaluator.Policy needs to be supported.
dataset = OPEDataset(...) # load your data, see details in `d3pe.utils.data`

evaluator = get_evaluator_by_name('fqe')() # create an ope evaluator, for example FQE
evaluator.initialize(dataset) # initialize the evaluator
ope_score = evaluator(policy) # evaluate the policy