PyTorch Finance Starter Code

This repository provides a starter code for a finance deep learning project using PyTorch.

The repository is organized as follows.

├── 📂 config/
│   ├── 📄 dvc_pipeline.yaml
│   ├── 📄 tune.yaml
│   ├── 📄 train.yaml
│   └── 📄 inference.yaml
├── 📂 dataset/
│   └── 📂 raw/
│       └── 📄 sample_data.csv
├── 📂 notebook/
│   ├── 📄 inference.ipynb
├── 📂 src/
│   ├── 📄 __init__.py
│   ├── 📄 data.py
│   ├── 📄 model.py
│   ├── 📄 utils.py
│   ├── 📄 tune.py
│   ├── 📄 train.py
│   ├── 📄 app.py
│   ├── 📄 serve.py
│   └── 📄 deploy.py
├── 📄 dvc.yaml
└── 📄 requirements.txt

• config contains the default YAML configuration files for the scripts.

  • config/dvc_pipeline.yaml contains the hyperparameters and configurations for the DVC pipeline in dvc.yaml.

  • config/tune.yaml contains the hyperparameters and configurations for src/tune.py.

  • config/train.yaml contains the hyperparameters and configurations for src/train.py.

  • config/inference.yaml contains the configurations for notebook/inference.ipynb, src/app.py, src/serve.py, and src/deploy.py.

• dataset is the directory for raw and processed data files (to be tracked by DVC).

  • dataset/raw/ is the directory for placing raw data files.

  • dataset/raw/sample_data.csv contains a sample PSE data.

• notebook contains the Jupyter notebooks.

  • notebook/inference.ipynb is the notebook for performing model inference and other analysis.

• src contains the Python scripts.

  • src/data.py contains data-related classes and functions.

  • src/model.py contains model-related classes and functions.

  • src/utils.py contains additional utility functions.

  • src/tune.py contains the code for hyperparameter tuning.

  • src/train.py contains the code for training the model.

  • src/app.py contains the code for an inference front-end for the model.

  • src/serve.py contains the code for locally serving the model.

  • src/deploy.py contains the code for deploying the model to Docker.

• dvc.yaml defines the DVC pipeline for hyperparameter tuning and model training.

• requirements.txt lists the Python package dependencies for this repository.

Installation

To install the dependencies of this repository, run the command

pip install -r requirements.txt

DVC Pipeline

To use the DVC pipeline, first remove the raw dataset from being git-tracked by running the command

git rm -r --cached dataset/raw/sample_data.csv
git commit -m 'Stopped Tracking Dataset'

To perform hyperparameter tuning and model training using the pipeline, edit the configuration config/dvc_pipeline.yaml and then run

dvc exp run

This will create params.yaml containing the final configurations used in the pipeline.

Hyperparameter Tuning

To perform hyperparameter tuning, edit the configuration config/tune.yaml to specify the search space and then run

python -m src.tune

Model Training

After tuning the hyperparameters, edit the configuration config/train.yaml with the best hyperparameter and then run

python -m src.train

Model Inference

To perform model inference, edit the configuration config/inference.yaml to specify the MLFlow and other settings.

Run the notebook notebook/inference.ipynb to perform model inference and other analysis.

Run the following to serve the model with a Gradio front-end app

python -m src.app

Run the following to serve the model locally with an API

python -m src.serve

Run the following to deploy the model with Docker

python -m src.deploy