You can deploy this model and start earning money today!

harukk11/
deliberation-controller-checkpoints

Name: deliberation-controller-checkpoints
Availability: InStock

Views

No views yet

Deliberation Controller

This repository contains data processing, controller modeling, and supervised training code for a Deliberation Controller.

Project Layout

deliberation_controller/data/: signal extraction, normalization, dataset preparation
deliberation_controller/model/: controller model definitions
deliberation_controller/train/: training scripts
legacy analysis scripts at repository root are kept for reference

Environment

bash

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

If you train on GPU, install a CUDA-compatible PyTorch build for your target server.

Prepare Dataset

bash

python -m deliberation_controller.data.prepare_dataset \
  --input "<path_to_trajectory_json>" \
  --output "./deliberation_controller/data/dataset.json" \
  --reference-distribution "./deliberation_controller/data/reference_distribution.json" \
  --window-size 5 \
  --seed 42 \
  --rebuild-reference

Train (Supervised)

bash

python -m deliberation_controller.train.train_sl \
  --data_path ./deliberation_controller/data/dataset.json \
  --epochs 100 \
  --batch_size 256 \
  --lr 1e-3 \
  --patience 10 \
  --save_dir ./deliberation_controller/checkpoints \
  --gate_threshold 0.5

Notes

action_label = -100 indicates gate=0 samples and is ignored in action loss.
Large datasets/checkpoints are ignored by default via .gitignore.

harukk11/deliberation-controller-checkpoints

Deliberation Controller

Project Layout

Environment

Prepare Dataset

Train (Supervised)

Notes

harukk11/
deliberation-controller-checkpoints