Estimating data quality is crucial problem in Offline Reinforcement Learning. Unlike in Online RL, data, collected from some agent has a strong impact on agent final performance. ODQM or Offline Data Quality Measurement provides methods for estimating provided data quallity

If you want to use D4RL offline envs, install it following instructions from official repository

First, you needed to clone this repository:

git clone https://github.com/zen0no/odqm

Then install requirements:

pip install -r requirements.txt

To estimate your data, firstly you need to create directory structure like in sample_data

sample_data
├── actions.npy
├── dones.npy
├── rewards.npy
└── states.npy

Next, replace data name from config/sample_config.yaml to your data directory and run:

python3 train.py config/sample.yaml

Optionally, you can write your own yaml config and provide path to it as an argument.