Data

To enhance extensibility and reuse, SATGL’s data module implements a processing pipeline to transform raw data into model-ready inputs.

The workflow functions as follows:

The details are as follows:

• Raw Input

  Unprocessed raw input dataset. Detailed as Raw Input.

• Dataset

  The Dataset is indeed implemented based on DGLDataset. It serves as a foundational structure for creating specific datasets related to SAT tasks. The dataset initializes parameters, constructs graphs based on model settings using the DGL library, and contains abstract methods to be implemented by subclasses for loading data and defining dataset behavior. This design ensures compatibility with DGL and allows for customization when creating datasets for SAT-related tasks. Detailed as Dataset.

• DataLoader

  The DataLoader module provides a convenient interface for loading and batching data from a custom dataset designed for SAT tasks. It utilizes DGL’s graph representation and extends it to handle various data types efficiently. The DataLoader module extends the capabilities of DGL for efficient handling of datasets during training. It inherits from both dgl.dataloading.GraphCollator and dgl.dataloading.GraphDataLoader in the DGL library. Detailed as DataLoader.

Here are the related docs for data module:

• Raw Input
• Dataset
• DataLoader