apowerb/th2forecast repository overview

⁠th2forecast Docker Image

th2Forecast Docker image is designed for automated time series forecasting and machine learning model evaluation. It provides an integrated environment that streamlines the complete forecasting pipeline, from data preparation to final prediction visualization, packaged for containerized deployment.

⁠Main Features

• Automated Preprocessing: Includes robust functions for cleaning time series data, handling missing values, detecting anomalies, and identifying level shifts.
• Advanced Feature Engineering: Specialized modules for generating lag features and integrating exogenous data (such as holidays or weather) to enhance machine learning model performance.
• Diverse Model Support: Predict using a wide array of engines, including:
  • ARIMA
  • Prophet
  • MARS
  • Linear Regression
  • Random Forest
  • XGBoost
• Model Tuning & Resampling: Optimize hyper-parameters and validate models using time-series cross-validation.
• Scalability: Support for distributed processing via Spark for large-scale forecasting tasks.
• Interactive UI: Comes with a built-in Shiny module for interactive data upload, model configuration, and performance visualization.

⁠Workflow

1. Data Ingestion: Upload your time series dataset.
2. Preprocessing: Clean data and resolve anomalies.
3. Feature Engineering: Generate predictive features optimized for ML algorithms.
4. Modeling: Select algorithms, tune parameters, and train models.
5. Forecasting: Generate forecasts and visualize predictive performance.

⁠Build

You can build the th2Forecast Docker image from the source:

docker build -t th2forecast:latest .

⁠Running the Container

To launch the containerized service:

docker run -d -p 8000:8000 --name th2forecast-api th2forecast:latest

⁠API Access

th2forecast includes an integrated Plumber API for remote forecasting tasks. For details on how to interact with the API, please refer to the documentation or the endpoint definitions in your deployment.

⁠License

GLP-3