# DMLab-30 for Human-like Agent

This repository is a modification for human-like agent from the SEED version `091e540541eaaedec31d5855059018b50ab11293`
The original source is from the open source SEED(https://github.com/google-research/seed_rl)

## How to use

- Clone the original SEED project

git clone https://github.com/google-research/seed_rl.git
cd seed_rl

- Replace the `learner.py` for v-trace agent and the `env.py` in `dmlab`

- For other commands, please follow SEED

## Problems
1. version not found.

2. ./run_local.sh not executable in NV.

# Unity ML-Agents for Human-like Agent

This repository is a modification for human-like agent from the `mlagents` (version tag: 0.8.1)
The original source is from the open source mlagents(https://github.com/Unity-Technologies/ml-agents)

## Install

- Install `mlagents.envs`
Please refer to https://github.com/Unity-Technologies/ml-agents/tree/0.8.1/ml-agents-envs

- Install `mlagents.trainers`
Example command:

pip install ../Codes/mlagents

## How to use

The corresponding enviroments are in `Games` folders

- Training
Example command:

mlagents-learn {trainer_config.yaml} --env="{YOUR_PATH}/GameForTrain/Unity Environment.exe" --run-id="{YOUR_MODEL}" --train 

About the configure file, the example file is in `Config`

- Evalution
Example command:

mlagents-learn {trainer_config.yaml} --env="{YOUR_PATH}/GameForEval/Unity Environment.exe" --run-id="{YOUR_MODEL}" --load

- Human play
Use `GameForHuman` for human playing

## My Installation

conda create -n mlagents-old python=3.6
conda activate mlagents-old
git clone https://github.com/Unity-Technologies/ml-agents -b 0.8.1
pip install mlagents-envs==0.8.1
cd Human-likeRL/Unity
pip install Codes/

mlagents-learn Config/trainer_config.yaml --env="Games/GameForTrain/Unity Environment.exe" --run-id="MyModel" --train

## Problems
1. Install mlagents.trainers error:



Solution:
copy README.md from origial 0.8.1 repo


2. Run training error:

mlagents-learn Config/trainer_config.yaml --env="Games/GameForTrain/Unity Environment.exe" --run-id="MyModel" --train