G-Tric

G-Tric generates synthetic three-dimensional datasets with configurable planted triclusters to evaluate and benchmark triclustering algorithms across domains such as biology, social sciences, urban studies, and geophysics.

Key Features:

• Synthetic Data Generation: Generates synthetic three-dimensional datasets with configurable properties to provide known ground truth for triclustering evaluation.
• Configurable Properties: Allows specification of data types (numeric or symbolic), dimensions, and background distributions.
• Planted Triclusters: Embeds triclusters with adjustable patterns, structures, and interactions (including overlapping) within the synthetic data.
• Data Quality Control: Introduces configurable levels of missing data, noise, or errors to assess algorithm robustness.
• Benchmark Datasets: Includes predefined benchmark datasets that replicate widely used three-way data and provide corresponding planted triclustering solutions and generating parameters.
• Evaluation Metrics: Supports intrinsic and extrinsic metrics for triclustering evaluation to enable comprehensive algorithm comparisons.

Scientific Applications:

• Algorithm benchmarking: Comparative evaluation of triclustering algorithms against known ground truth.
• Robustness assessment: Testing algorithm performance under varying levels of missing data, noise, and errors.
• Method development for domain-specific data: Developing and validating triclustering methods for three-way data in biology, social sciences, urban studies, and geophysics.

Methodology:

Generates three-dimensional datasets by configuring data types (numeric or symbolic), dimensions, and background distributions; embeds planted triclusters with specified patterns, structures, and overlaps; injects missing data, noise, or errors; and records planted solutions and generating parameters.