🌊 StreamFlow-ML

A modular machine learning framework for streamflow prediction using deep learning architectures such as LSTM, BiLSTM, and GRU. The framework supports flexible configuration, training, hyperparameter tuning, and evaluation of hydrologic time-series models.

---

📁 Project Structure

StreamFlow-ML/
├── config/             # YAML configuration files
├── data/               # Input data (not committed due to size)
├── hpt/                # Hyperparameter tuning scripts/configs
├── models/             # Deep learning model architectures
│   ├── bilstm_model.py
│   ├── gru_model.py
│   └── lstm_model.py
├── notebooks/          # Jupyter notebooks for experiments/visuals
├── results/            # Evaluation results and metrics
├── saved_models/       # Saved trained model weights
├── utils/              # Utilities (data loading, evaluation)
│   ├── data_loader.py
│   └── evaluation.py
├── train.py            # Script to train models
├── tune_model.py       # Script for hyperparameter tuning
├── predict.py          # Script for prediction/inference
└── README.md           # Project documentation

---

⚙️ Setup

Make sure Python ≥ 3.7 is installed. Then:

pip install -r requirements.txt

---

🚂 Training

python train.py

You can also customize the training process by editing YAML configs under config/.

---

🔍 Evaluation

python -m utils.evaluation

This runs model evaluation (e.g., NSE, RMSE, Kling-Gupta) and logs results into the results/ folder.

---

🧪 Hyperparameter Tuning

python tune_model.py

This runs your HPT strategy defined in the hpt/ folder.

---

📈 Models Supported

• LSTM
• BiLSTM
• ConvID
• GRU
  (Architectures are located in models/)

---

📌 Notes

• Large .csv/dataset files were removed from GitHub due to file size limits.
• Compatible with site-specific and multi-site streamflow prediction.
• All models are configurable via external YAML files.

---

🙋‍♂️ Maintainer

Developed and maintained by @Almehedi06

---

📜 License

MIT License