Created base structure for adding discovery and other algorithms

[amit-sharma]

New files added that convey the directory structure.

This PR does not add any functionality, just the protocol class and structure. The goal is to enable collaborators to contribute code.

[Copilot]

Pull Request Overview

This PR establishes the foundational structure for a causal discovery library by creating base interfaces and utility functions. The changes introduce a protocol-based design for datasets and skeleton implementations for evaluation metrics.

• Defines a Dataset protocol with methods for graph access, data retrieval, and synthetic data generation
• Creates placeholder functions for standard causal discovery evaluation metrics (accuracy, precision, recall, F1)

Reviewed Changes

Copilot reviewed 2 out of 8 changed files in this pull request and generated 4 comments.

File	Description
pywhyllm/datasets/dataset.py	Defines the Dataset protocol interface with methods for graph, data, and synthetic data generation
pywhyllm/datasets/metrics.py	Creates skeleton functions for evaluation metrics used in causal discovery

Comments suppressed due to low confidence (1)

pywhyllm/datasets/metrics.py:21

• Function name 'F1' should follow Python naming conventions. Consider renaming to 'f1_score' or 'f1' (lowercase).

def F1(edges, true_edges)

[emrekiciman]

This is great. I added a few questions about the structure.

Also: do we want this dataset protocol to integrate well with data handling in other pywhyllm libraries? Or is this only for self-contained pywhy-llm benchmarking for example?

[emrekiciman]

Is this supposed to return the ground truth graph, or is this function intended to execute the PyWhyLLM functions to derive a candidate graph?

[emrekiciman]

What do you think about merging data() and generate_data() into a single function? Then some Dataset objects might be synthetic Datasets, and some might be grounded datasets (real data)?

[emrekiciman]

Given edges and true_edges, is the calculation of accuracy, precision, and recall specific to the dataset?