This repository contains the code for my Machine Learning Algorithms and Applications Project.

Installation

The simplest way to install the required packages is to use the provided environment.yml file and conda:

conda env create -f environment.yml
conda activate wlclf

Once installed you can experiment with the code in the notebooks.

Functionality

Currently, two main functionalities are implemented, the representation and evaluation of C2 formulas and a simple graph classifier based on color refinement and decision trees.

C2 Formulas

In c2.py the class Formula is used to represent and evaluate formulas in the logic C2, two-variable first order logic with counting quantifiers. The formulas can be evaluated on graphs:

from c2 import *
import networkx as nx

formula = Exists(Var.x, (Exists(Var.y, E(Var.x, Var.y), 7)))
formulas.evaluate(nx.fast_gnp_random_graph(10, 0.5))

Graph Classification

In wlclassifier.py a simple graph classifier is implemented:

phi = Exists(Var.x, (Exists(Var.y, E(Var.x, Var.y), 7)))
X = [nx.fast_gnp_random_graph(10, 0.5) for _ in range(100)]
y = [phi.evaluate(g) for g in X]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

clf = WLClassifier()
clf.fit(X_train, y_train)
clf.score(X_test, y_test)