mirror of
https://github.com/kuhyx/WUT_Computer_Science.git
synced 2026-07-04 20:23:04 +02:00
Krzysztof kuhy Rudnicki
9576b25315
git-subtree-dir: Programming/USD git-subtree-mainline:7cf25fc048git-subtree-split:b10db883f2
48 lines
1.1 KiB
Markdown
48 lines
1.1 KiB
Markdown
# USD
|
|
|
|
## Task:
|
|
Zapoznaj się z [MetaDrive](https://github.com/metadriverse/metadrive/). Wytrenuj co najmniej dwóch różnych agentów
|
|
wykorzystując algorytmy **wieloagentowe** (MA), na co
|
|
najmniej trzech różnych mapach. Omów otrzymane wyniki oraz zwizualizuj działanie
|
|
wytrenowanych agentów.
|
|
|
|
|
|
## Setup
|
|
```
|
|
conda create -n copo python=3.7
|
|
conda activate copo
|
|
|
|
# Install MetaDrive version 0.2.5
|
|
pip install git+https://github.com/metadriverse/metadrive.git@releases/0.2.5
|
|
|
|
pip install torch
|
|
|
|
git clone https://github.com/decisionforce/CoPO
|
|
cd CoPO/copo_code
|
|
pip install -e .
|
|
|
|
pip install -U ray==1.2.0 "ray[rllib]==1.2.0"
|
|
pip install -U "numpy<1.19.0"
|
|
pip uninstall opencv-python
|
|
pip uninstall opencv-python-headless
|
|
pip install opencv-python==4.5.5.64
|
|
pip install pydantic==1.9.0
|
|
```
|
|
|
|
## How to train a RL agents
|
|
|
|
```
|
|
cd CoPo/copo_code/copo/
|
|
python train_all_cl.py --exp-name my_cl
|
|
```
|
|
Training process 4.7h
|
|
```
|
|
python train_all_ippo.py --exp-name my_ippo
|
|
```
|
|
Training process 7.3h
|
|
|
|
## How to evaluate
|
|
```
|
|
python eval.py --root my_cl
|
|
python eval.py --root my_ippo
|
|
``` |