lagom is a light PyTorch infrastructure to quickly prototype reinforcement learning algorithms. Lagom is a 'magic' word in Swedish, "inte för mycket och inte för lite, enkelhet är bäst", meaning "not too much and not too little, simplicity is often the best". We use this name because lagom is also the philosophy how this library is designed and inspired.


lagom balances between the flexibility and the userability when developing reinforcement learning (RL) algorithms. The library is built on top of PyTorch and provides modular tools to quickly prototype RL algorithms. However, we do not go overboard, because in practice, going too low level is rather time consuming and prone to potential bugs and going too high level degrades the flexibility which making it difficulty to try out some crazy ideas.

We shall continuously try to make lagom to be more 'self-contained' to run experiments quickly. Now, it internally supports base classes for multiprocessing (master-worker framework) to parallelize (e.g. experiments and evolution strategies). It also supports hyperparameter search by defining configurations either as grid search or random search.

One of the main pipelines to use lagom can be done as following:

  1. Define environment and RL agent
  2. User runner to collect data for agent
  3. Define algorithm to train agent
  4. Define experiment and configurations.

A graphical illustration is coming soon.


Install dependencies

This repository requires following packages:

  • Python >= 3.6
  • pytest >= 3.6.3
  • setuptools >= 39.0.1
  • Numpy >= 1.14.5
  • Matplotlib >= 2.2.2
  • PyTorch >= 0.5.0a0
  • gym >= 0.10.5
  • cma >= 2.6.0

There are bash scripts in scripts/ directory to automatically set up the conda environment and dependencies.

Install lagom

git clone
cd lagom
pip install -e .

Getting Started

Detailed tutorials is coming soon. For now, it is recommended to have a look in examples/ or source code.


We shall continuously provide examples/ to use lagom.


We are using pytest for tests. Feel free to run via

pytest test -v



- Readthedocs Documentation
- Tutorials

More standard RL baselines

- Q-Prop
- DQN: Rainbow

More standard networks

- Monte Carlo Dropout/Concrete Dropout


- VecEnv: similar to that of OpenAI baseline
- Support pip install
- Technical report


This repo is inspired by OpenAI rllab, OpenAI baselines, RLPyTorch, TensorForce, and Intel Coach

Please use this bibtex if you want to cite this repository in your publications:

      author = {Zuo, Xingdong},
      title = {lagom: A light PyTorch infrastructure to quickly prototype reinforcement learning algorithms},
      year = {2018},
      publisher = {GitHub},
      journal = {GitHub repository},
      howpublished = {\url{}},