Skip to content

CONFIG_GUIDE.md

Latest commit

 

History

History
111 lines (78 loc) · 2.55 KB

CONFIG_GUIDE.md

File metadata and controls

111 lines (78 loc) · 2.55 KB

Configuration Guide

This project uses YAML config files to manage training and inference parameters instead of command-line arguments.

Quick Start

Training

  1. Create a config file based on config.yaml:
cp config.yaml my_experiment.yaml

  1. Edit my_experiment.yaml with your desired parameters

  2. Run training:

python train.py --config my_experiment.yaml

Inference

  1. Create a config file based on config_inference.yaml:
cp config_inference.yaml my_inference.yaml

  1. Edit my_inference.yaml (especially set ckpt_path to your model checkpoint)

  2. Run inference:

python inference.py --config my_inference.yaml

Config File Format

Config files are YAML format. Example:

# Dataset
dataset: "atlas"
protein: "1bx7"
pkl_path: "path/to/graphs.pkl"

# Model hyperparameters
latent_dim: 128
hidden_dim: 256
embedding_dim: 128

# Training
batch_size: 32
n_epochs: 500
lr: 1e-3

# ... more parameters

Command-Line Overrides

You can override any config parameter from the command line without editing the config file:

# Override single parameters
python train.py --config config.yaml --batch_size 64 --n_epochs 1000

# Override multiple parameters
python train.py --config config.yaml --latent_dim 256 --hidden_dim 512 --lr 5e-4

Utility Functions

The config system is implemented in utils/config.py with three main functions:

load_config(config_path)

Loads a config file (YAML or JSON) and returns a dictionary.

from utils.config import load_config
config = load_config("config.yaml")

config_to_hparams(config)

Converts a config dictionary to argparse.Namespace for easy attribute access.

from utils.config import config_to_hparams
args = config_to_hparams(config)
print(args.batch_size)  # Access as attribute

save_config(config, output_path, format='yaml')

Saves a config dictionary to a file (useful for recording training configs).

from utils.config import save_config
save_config(config, "saved_config.yaml")

Config Files Included

  • config.yaml: Default training config with typical hyperparameters
  • config_inference.yaml: Default inference config

Create experiment-specific config files by copying and modifying these templates.

Benefits

  • Reproducibility: Easily save and share configs used for experiments
  • Organization: Group related parameters by section
  • Flexibility: Mix config file defaults with command-line overrides
  • Version Control: Track config changes alongside code changes