Sovereign 8s.is mirror of standardgalactic/MARINA
  • Python 99.7%
  • Shell 0.3%
Find a file
Repository files (latest commit first)
Filename Latest commit message Latest commit date
2026-03-09 20:23:09 +11:00
factors_mining Initial commit 2026-03-09 20:10:19 +11:00
figures Initial commit 2026-03-09 20:10:19 +11:00
forecasting Initial commit 2026-03-09 20:10:19 +11:00
.gitignore Initial commit 2026-03-09 20:10:19 +11:00
README.md Update README 2026-03-09 20:23:09 +11:00
requirements.txt Initial commit 2026-03-09 20:10:19 +11:00

MARINA

This repository contains the implementation of MARINA, A Lightweight Physics-Aware Framework for Multi-Scale Marine Heatwaves Forecasting.

MARINA model architecture

  1. Stage1: Factor Mining (factors_mining/)
  2. Stage2: Forecasting (forecasting/)

Environment

Recommended:

  • Linux
  • CUDA-capable GPU
  • Python 3.8+ (3.8/3.9 recommended)

Install dependencies:

python -m pip install -r requirements.txt

Stage 1: Factor Mining

Data

Place factor-mining data under:

factors_mining/dataset/qilb_data/

The current repository already contains a factors_mining/dataset/ tree. If you use your own data, keep the same folder format.

Download the dataset from google driver, and place them in ./factors_mining/dataset/qilb_data folder.

Run

cd factors_mining

Daily:

bash scripts/daily/pool20/run.sh

Hourly:

bash scripts/hourly/pool20/run.sh

Weekly:

bash scripts/weekly/pool20/run.sh

Notes:

  • Different pool sizes are available in scripts/*/pool40, pool100, pool150, pool200.
  • Some scripts may contain hardcoded GPU IDs; adjust CUDA_VISIBLE_DEVICES if needed.

Stage 2: Forecasting

Data

Forecasting data should be under:

forecasting/datasets/

Download the dataset from google driver, and place them in ./forecasting/datasets folder.

Run with script

cd forecasting
bash script/run_nas_daily.sh

Run with CLI

cd forecasting
export OPENAI_API_KEY=your_api_key
CUDA_VISIBLE_DEVICES=0 python run_nas_train.py \
  --search_strategy llm \
  --llm_model gpt-4-1106-preview \
  --nas_trials 10

Key Forecasting Arguments

  • --freq: time granularity (d, h, w)
  • --station: station name (for example ARN)
  • --root_path, --data_path: forecasting dataset path
  • --seq_len, --pred_len, --label_len: sequence setup
  • --batch_size, --train_epochs, --learning_rate: training setup
  • --nas_trials, --llm_model, --llm_temperature: NAS/LLM setup
  • --nas_save_dir, --history_dir: output locations

Outputs

Typical forecasting outputs:

  • forecasting/nas/.../nas_results.json
  • forecasting/nas/.../nas_results.csv
  • forecasting/history/nas_marina_<freq>.json
  • forecasting/history/nas_marina_performance_<freq>.json