https://github.com/callmequant/financial_ts_generation_hackathon

https://github.com/callmequant/financial_ts_generation_hackathon

Last synced: about 13 hours ago
JSON representation

• Host: GitHub
• URL: https://github.com/callmequant/financial_ts_generation_hackathon
• Owner: CallmeQuant
• Created: 2024-06-10T15:04:22.000Z (5 months ago)
• Default Branch: main
• Last Pushed: 2024-06-10T15:16:53.000Z (5 months ago)
• Last Synced: 2024-06-10T17:28:16.156Z (5 months ago)
• Language: Python
• Size: 17.6 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# FourierFlow Generator for Financial Time Series

The FourierFlow is a simple, yet powerful tool for generating synthetic time series data using deep generative models. The model is slightly modified and applied to the Market Scenario Generator Hackathon: From Stability to Storms (for more details of the hackathon, visit [Hackathon Website](https://hackathon.deepintomlf.ai/competitions/38)).

# Getting Started

## Installation:

Clone this repository to your local machine.

I recommend to create a new environment by running

```console

conda env create -f environment.yml

```

## Configuration:

- Navigate to the config_ff.yaml file.

- Adjust the parameters according to your needs:

    + `input_dim`: Dimensionality of input features.

    + `output_dim`: Dimensionality of output features.

    + `hidden_dim`: Size of hidden layers.

    + `n_flows`: Number of flow layers.

    + `n_lags`: Number of time lags.

    + `vol_activation`: Activation function for volatility modeling (e.g., “softplus”).

    + Other hyperparameters (batch size, learning rate, etc.).

- Pretraining or Checkpoint:

    + Choose between two modes:

        + Pretrain:

            + Load your training data and labels (regular and crisis data).

            + Train the generator using generator_regular.fit() and generator_crisis.fit().

            + Save the combined model dictionary using save_combined_model_dict().

        + Checkpoint (Pickle Files):

            + Load the pre-trained model from model_dict.pkl.

- Generating Samples:

    + Run the script main.py in console:

        ```console

        python main.py

        ```

    + Specify the condition (e.g., crisis or regular) by setting `condition[0]` in the `main.py`.

    + The generated synthetic data will be saved to a pickle file.

# References:

[1] Alaa, A.M., Chan, A.J., & Schaar, M.V. (2021). Generative Time-series Modeling with Fourier Flows. International Conference on Learning Representations.