steveazzolin / advanced_machine_learning

Final project for the Advance Machine Learning course

Advanced Topics in Machine Learning and Optimization course project

This project is the final course project for the aforementioned course @ UNITN.

A detailed description of the project can be found in Report.pdf

Requisites

• Create a tmp folder in the root directory
• Import the conda env env.yml
• Install this modified version of DIG

Train a GNN

For example, to train a GCN over the CORA dataset:

cd Models
python models_CORA.py --model=GCN --train --save

By specifying also --wandb the logging to wanb is enabled

To run a trained model instead:

cd Models
python models_CORA.py --model=GCN

Extract explanation

python extract_explanations.py --model=GCN --dataset=CORA --expl=PGExplainer --save

Pattern mining

To extract the features of the local explanations, cutting irrelevant edges and removing the connected components not including the target node to be explained:

python mine_explanations.py --model=GCN --dataset=CORA --expl=PGExplainer --cut_edges --cut_cc

The code above can also be used to plot the local explanations, to plot the prototypes as found by the greedy Edit distance based algorithm, and to plot the edge scores distribution.