Metrics

This notebook shows the various classification metrics of scikit-network.

from IPython.display import SVG

import numpy as np

from sknetwork.data import art_philo_science
from sknetwork.classification import get_accuracy_score, get_f1_scores, get_average_f1_score, get_confusion_matrix
from sknetwork.classification import DiffusionClassifier
from sknetwork.visualization import svg_graph

Node classification

graph = art_philo_science(metadata=True)
adjacency = graph.adjacency
position = graph.position
names = graph.names
labels = graph.labels

# Visualization
image = svg_graph(adjacency, position=position, names=names, labels=labels)
SVG(image)

# Number of labels
n_labels = len(set(labels))

# Classification by diffusion
diffusion = DiffusionClassifier()

# Prediction
labels_pred = diffusion.fit_predict(adjacency, labels={i: labels[i] for i in [0, 10, 15, 20]})

Accuracy

accuracy = get_accuracy_score(labels, labels_pred)

np.round(accuracy, 2)

0.83

F1 scores

# all f1 scores
f1_scores = get_f1_scores(labels, labels_pred)

np.round(f1_scores, 2)

array([0.71, 1.  , 0.78])

# with precisions and recalls
f1_scores, precisions, recalls = get_f1_scores(labels, labels_pred, return_precision_recall=True)

np.round(precisions, 2)

array([0.86, 1.  , 0.69])

np.round(recalls, 2)

array([0.6, 1. , 0.9])

# average
f1_score = get_average_f1_score(labels, labels_pred)

np.round(f1_score, 2)

0.83

Confusion matrix

confusion = get_confusion_matrix(labels, labels_pred)

# True labels on rows
confusion.toarray()

array([[ 6,  0,  4],
       [ 0, 10,  0],
       [ 1,  0,  9]])