Skip to content
All library documents

A Configurable Neural Network Design for Time-Series and Tabular Data

Article Qlib

Summary

GeneralPtNN is presented as a redesign intended to support both time-series and tabular datasets through a common PyTorch workflow. The stated approach is to keep the workflow configurable and change the network and dataset classes when moving between data types. Example configurations demonstrate a GRU model for time-series input, an MLP configuration, and a conversion from a time-series setup to tabular data with limited configuration changes.

The document says the GRU configuration aligns with earlier results and that the MLP configuration provides similar functionality, but it supplies no quantitative results in the excerpt. It also notes that MLP results differ from an earlier implementation because training stops according to epochs here, while the prior method used a maximum number of steps. Existing models are still slated for alignment, so compatibility and result equivalence remain limited by implementation and stopping-rule differences.

Key ideas

  • GeneralPtNN aims to handle time-series and tabular data through configurable PyTorch models.
  • Changing the network and dataset classes is presented as the main conversion between data types.
  • The examples include GRU and MLP configurations.
  • Training stops by epoch in this design, unlike the earlier maximum-step method.
  • The excerpt gives no quantitative evidence and notes that model alignment remains incomplete.

Tags

Machine learningBacktestingStatistics
Full text
# Introduction




# Introduction

What is GeneralPtNN
- Fix previous design that fail to support both Time-series and tabular data
- Now you can just replace the Pytorch model structure to run a NN model.

We provide an example to demonstrate the effectiveness of the current design.
- `workflow_config_gru.yaml` align with previous results [GRU(Kyunghyun Cho, et al.)](../README.md#Alpha158-dataset)
  - `workflow_config_gru2mlp.yaml` to demonstrate we can convert config from time-series to tabular data with minimal changes
    - You only have to change the net & dataset class to make the conversion.
- `workflow_config_mlp.yaml` achieved similar functionality with [MLP](../README.md#Alpha158-dataset)

# TODO

- We will align existing models to current design.

- The result of `workflow_config_mlp.yaml` is different with the result of [MLP](../README.md#Alpha158-dataset) since GeneralPtNN has a different stopping method compared to previous implementations. Specificly, GeneralPtNN controls training according to epoches, whereas previous methods controlled by max_steps.

Shown in full with attribution under the source's licence. Licence: MIT

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.