Basic

Installation

Note: To use the baselines, you need Git LFS extension installed so you can pull the binaries from the repo.

To use Relezoo, you must clone the repo and install it via Make:

git clone https://github.com/Ohtar10/rele-zoo.git
cd rele-zoo
make install-env
make install
conda activate rele-zoo

Or directly via pip:

git clone https://github.com/Ohtar10/rele-zoo.git
cd rele-zoo
conda env create -f environment.yaml
conda activate
pip install .

If you want to use the baselines and you cloned the repository before installing Git LFS, after installing it, simply run git lfs fetch or git lfs pull inside local repository to fetch the actual binaries.

Verify Installation

To check if the installation works works good, first check the relezoo-run command, and then execute the default run.

relezoo-run --help
== Relezoo-run ==

This is Relezoo-run!
You can change different experiment
configuration groups by appending
== Configuration groups ==
Compose your configuration from those groups (algorithm=reinforce)
By default, relezoo will train a REINFORCE algorithm against cartpole environment,
so by just invoking ``relezoo-run`` you can check everything is working fine...

By default, relezoo is configured to run a reinforce algorithm on a single cartpole environment, so you can just run the program without any arguments:

relezoo-run
[2022-05-10 20:49:01,297][relezoo.engine.runner][INFO] - Running Training Session with config:
[2022-05-10 20:49:01,300][relezoo.engine.runner][INFO] - ...
[2022-05-10 20:49:01,301][relezoo.engine.runner][INFO] - Press CTRL + C to cancel the run
[2022-05-10 20:49:01,301][relezoo.engine.runner][INFO] - A checkpoint will be saved automatically after a successful run or cancel.
100%|████████████████████████████████████████████████████| 50/50 [00:19<00:00,  2.62it/s, loss=3.35, mean_batch_score=84.50, mean_batch_ep_length=84.50, mean_reward_100=32.82]
[2022-05-10 20:49:22,318][relezoo.engine.runner][INFO] - Training finished -- Mean reward over 100 epochs: 32.82
[2022-05-10 20:49:22,318][relezoo.engine.runner][INFO] - Saving current progress...

If you see a similar output, everything is working fine.

Development Mode Install

For development and testing, i.e., developing new models or experiments and running the test suites you can install in dev mode:

git clone https://github.com/Ohtar10/rele-zoo.git
cd rele-zoo
make install-env
make install-dev
conda activate rele-zoo

Or directly via pip:

git clone https://github.com/Ohtar10/rele-zoo.git
cd rele-zoo
conda env create -f environment.yaml
conda activate
pip install ".[dev]"