💡 The Overall Design of BasicTS

In a time series forecasting pipeline, the following key components are typically involved:

• Dataset: Specifies the methods for reading datasets and generating samples. (Located in basicts.data)
• Scaler: Manages the normalization and denormalization of datasets, supporting techniques such as Z-score and Min-Max normalization. (Located in basicts.scaler)
• Metrics: Defines the evaluation metrics and loss functions, including MAE, MSE, MAPE, RMSE, and WAPE. (Located in basicts.metrics)
• Runner: The core module of BasicTS, responsible for orchestrating the entire training process. The Runner integrates components such as Dataset, Scaler, Model, and Metrics, and provides a wide range of features. (Located in basicts.runner)
• Model: Defines the model architecture and the forward propagation process. (Located in basicts.models)

BasicTS includes a comprehensive and extensible set of components, enabling users to complete most tasks simply by providing a model structure.

To streamline the configuration of training strategies and centralize all options for easy and fair comparisons, BasicTS follows an all-in-configuration design philosophy.

Users can easily configure models, datasets, scaling methods, evaluation metrics, optimizers, learning rates, and other hyperparameters by modifying the configuration file—as simple as filling out a form.

🧑‍💻 Explore Further

• 🎉 Getting Stared
• 💡 Understanding the Overall Design Convention of BasicTS
• 📦 Exploring the Dataset Convention and Customizing Your Own Dataset
• 🛠️ Navigating The Scaler Convention and Designing Your Own Scaler
• 🧠 Diving into the Model Convention and Creating Your Own Model
• 📉 Examining the Metrics Convention and Developing Your Own Loss & Metrics
• 🏃‍♂️ Mastering The Runner Convention and Building Your Own Runner
• 📜 Interpreting the Config File Convention and Customizing Your Configuration
• 🎯 Exploring Time Series Classification with BasicTS
• 🔍 Exploring a Variety of Baseline Models