Tibo De Peuter
f31436bccd
- Add default_ppo.yaml template for training configurations - Create configs/README.md with usage documentation - Enable per-researcher configuration without code changes - Support YAML config files with CLI parameter overrides - Document how to set personal WandB credentials safely This allows researchers to maintain personal configs without committing credentials to the repository.
48 lines
1.3 KiB
Markdown
48 lines
1.3 KiB
Markdown
# Configuration Files
|
|
|
|
This directory contains configuration files for training experiments.
|
|
|
|
## Usage
|
|
|
|
Configuration files use YAML format and allow you to specify all training parameters in one place.
|
|
|
|
### Quick Start
|
|
|
|
Copy the default configuration template:
|
|
```bash
|
|
cp configs/default_ppo.yaml configs/my_experiment.yaml
|
|
```
|
|
|
|
Edit `my_experiment.yaml` to customize your experiment settings, particularly:
|
|
- `wandb_entity`: Your WandB username or team name
|
|
- `track`: Set to `true` to enable WandB logging
|
|
- Training hyperparameters as needed
|
|
|
|
Run training with your config:
|
|
```bash
|
|
python src/train.py --config configs/my_experiment.yaml
|
|
```
|
|
|
|
### Override Parameters
|
|
|
|
You can override any parameter from the command line:
|
|
```bash
|
|
python src/train.py --config configs/my_experiment.yaml --learning-rate 0.001 --num-envs 32
|
|
```
|
|
|
|
### Configuration for Different Users
|
|
|
|
Each researcher should create their own config file with their WandB settings:
|
|
```yaml
|
|
# configs/researcher_name.yaml
|
|
exp_name: "researcher_name_experiment"
|
|
track: true
|
|
wandb_project_name: "PPO-Modularity"
|
|
wandb_entity: "your-wandb-username" # Change this!
|
|
```
|
|
|
|
This approach allows everyone to use the codebase without modifying source files.
|
|
|
|
## Available Configurations
|
|
|
|
- `default_ppo.yaml` - Default PPO training configuration template
|