Skip to content

Latest commit

 

History

History
93 lines (66 loc) · 3.19 KB

README.md

File metadata and controls

93 lines (66 loc) · 3.19 KB

TimeSeriesModelingToolkit (TSMT)

A versatile toolkit for analyzing, preprocessing, and modeling time-series data. This framework provides tools for data scaling, filtering, sequence generation, and training various LSTM models, including both standard and TimeDistributed approaches.

Key Features

  • Comprehensive Data Processing: Scale, filter, and prepare time-series data for supervised learning.
  • Flexible Model Architectures: Choose between standard and TimeDistributed LSTM models, with advanced configurations including attention mechanisms and residual connections.
  • Batch Data Generation: Efficiently load data in batches for large datasets to optimize memory usage.
  • Model Training and Evaluation: Use built-in training configurations with callbacks for early stopping, learning rate reduction, and model checkpointing.

Repository Structure

  • dataset.py: Contains the Dataset class for data preprocessing, scaling, and sequence generation for model training.
  • data_generator.py: Manages data batching, enabling efficient loading for large time-series datasets.
  • LSTMRegressor.py: Implements various LSTM models, including both standard and TimeDistributed architectures.
  • model.py: Defines the ModelTrainer class, which supports early stopping, model checkpointing, and learning rate adjustments.
  • eval_model.py: Provides evaluaion functions to assess model performance, including metrics such as MAE and Pearson’s correlation.

Setup

Requirements

Install the required packages by running:

pip install -r requirements.txt

Folder Structure

To organize saved models, scalers, and data files, follow this folder structure:

TimeSeriesModelingToolkit/
├── dataset.py
├── data_generator.py
├── LSTMRegressor.py
├── model.py
├── eval_model.py
├── saved_model/
├── scalers/
└── data/

Setup

Integration with a New Project

To integrate TSMT into your project, clone the repository into your project directory:

git clone [email protected]:mikjkd/TSMT.git TSMT

Organize your project directory as follows:

forecast_project/
├── main.py
├── TSMT/
├── data/
├── saved_model/
├── scalers/
└── encoders/

Install the required packages by running:

pip install -r TSMT/requirements.txt

Usage

  1. Place your data in the data/ folder.
  2. Write the main.py following the main_example.txt
  3. Configure main.py with your dataset and desired settings.
  4. Run the training script:
python main.py

  1. Evaluation: After training, main.py will save evaluation metrics and generate visualizations. The eval_model.py script allows customization of evaluation metrics.

  2. Model Customization: LSTMRegressor.py provides different LSTM configurations, including both standard and TimeDistributed options with advanced features like attention.

Future Improvements

  • Experiment with different attention mechanisms for improved accuracy.
  • Extend model options with additional sequence models (e.g., GRU, Transformer).
  • Explore hyperparameter tuning for dataset-specific performance optimization.