From f31436bccd14fdb7866f6d7a826167fcb23257a5 Mon Sep 17 00:00:00 2001
From: Tibo De Peuter <tibo@depeuter.dev>
Date: Tue, 31 Mar 2026 19:53:15 +0000
Subject: [PATCH] feat(config): add YAML configuration system

- 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.
---
 configs/README.md        | 48 ++++++++++++++++++++++++++++++++++++++++
 configs/default_ppo.yaml | 48 ++++++++++++++++++++++++++++++++++++++++
 2 files changed, 96 insertions(+)
 create mode 100644 configs/README.md
 create mode 100644 configs/default_ppo.yaml

diff --git a/configs/README.md b/configs/README.md
new file mode 100644
index 0000000..d06fbb2
--- /dev/null
+++ b/configs/README.md
@@ -0,0 +1,48 @@
+# 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
diff --git a/configs/default_ppo.yaml b/configs/default_ppo.yaml
new file mode 100644
index 0000000..5775a84
--- /dev/null
+++ b/configs/default_ppo.yaml
@@ -0,0 +1,48 @@
+# PPO Training Configuration Template
+#
+# This file provides an example configuration for PPO training.
+# Copy this file and modify it for your specific experiments.
+#
+# Usage:
+#   python src/train.py --config-path configs/my_config.yaml
+# Or override specific parameters:
+#   python src/train.py --learning-rate 0.001 --num-envs 32
+
+# Experiment settings
+exp_name: "brittle_star_ppo"
+seed: 1
+
+# Tracking settings
+track: false  # Set to true to enable WandB logging
+wandb_project_name: "PPO-Modularity"
+wandb_entity: null  # Set to your WandB username or team name
+
+# Model saving
+save_model: true
+checkpoint_frequency: 100  # Save checkpoint every N iterations (0 = no checkpoints)
+
+# Environment settings
+num_envs: 16
+
+# Training hyperparameters
+total_timesteps: 10000000
+learning_rate: 0.00025
+num_steps: 128
+anneal_lr: true
+
+# PPO specific
+gamma: 0.99
+gae_lambda: 0.95
+num_minibatches: 4
+update_epochs: 4
+norm_adv: true
+clip_coef: 0.1
+clip_vloss: true
+ent_coef: 0.01
+vf_coef: 0.5
+max_grad_norm: 0.5
+target_kl: null
+
+# Hardware
+cuda: true
+torch_deterministic: true