Feature ranking with self-attention networks

This is the repository of the SAN paper, found here:

@inproceedings{DBLP:conf/ecai/SkrljDLP20,
  author    = {Blaz Skrlj and
               Saso Dzeroski and
               Nada Lavrac and
               Matej Petkovic},
  editor    = {Giuseppe De Giacomo and
               Alejandro Catal{\'{a}} and
               Bistra Dilkina and
               Michela Milano and
               Sen{\'{e}}n Barro and
               Alberto Bugar{\'{\i}}n and
               J{\'{e}}r{\^{o}}me Lang},
  title     = {Feature Importance Estimation with Self-Attention Networks},
  booktitle = {{ECAI} 2020 - 24th European Conference on Artificial Intelligence,
               29 August-8 September 2020, Santiago de Compostela, Spain, August
               29 - September 8, 2020 - Including 10th Conference on Prestigious
               Applications of Artificial Intelligence {(PAIS} 2020)},
  series    = {Frontiers in Artificial Intelligence and Applications},
  volume    = {325},
  pages     = {1491--1498},
  publisher = {{IOS} Press},
  year      = {2020},
  url       = {https://doi.org/10.3233/FAIA200256},
  doi       = {10.3233/FAIA200256},
  timestamp = {Tue, 15 Sep 2020 15:08:42 +0200},
  biburl    = {https://dblp.org/rec/conf/ecai/SkrljDLP20.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

http://ecai2020.eu/papers/1721_paper.pdf (please, cite if you are using it!). Note that the full code with datasets to reproduce the paper can be found here: https://gitlab.com/skblaz/attentionrank (the code is in benchmark-ready form). The purpose of this repository is to provide all functionality in a user-friendly way. Disclaimer: this code was not extensively benchmarked and can contain bugs. If you find one, please open an issue.

Installing SANs

python setup.py install

pip install git+https://github.com/SkBlaz/san

Using SANs

A simple usecase is given next:

from scipy import sparse
import numpy as np
from sklearn.datasets import load_breast_cancer
import san
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.feature_selection import chi2,f_classif,mutual_info_classif
from sklearn.ensemble import RandomForestClassifier

sns.set_style("whitegrid")
dataobj = load_breast_cancer()
X = dataobj['data']
Y = dataobj['target']
names = dataobj['feature_names']

# let's overfit, just for demo purposes
clf = san.SAN(num_epochs = 32, num_heads = 2, batch_size = 8, dropout = 0.2, hidden_layer_size = 32)
X = sparse.csr_matrix(X)
clf.fit(X, Y)
preds = clf.predict(X)
global_attention_weights = clf.get_mean_attention_weights()
local_attention_matrix = clf.get_instance_attention(X)
mutual_information = mutual_info_classif(X,Y)
rf = RandomForestClassifier().fit(X,Y).feature_importances_

plt.plot(names, global_attention_weights, label = "Global attention", marker = "x")
plt.plot(names, np.mean(local_attention_matrix, axis = 0), label = "Local attention - mean", marker = "x")

plt.plot(names, np.max(local_attention_matrix, axis = 0), label = "Local attention - max", marker = "x")

plt.plot(names, mutual_information, label = "Mutual information", marker = ".")

plt.plot(names, rf, label = "RandomForest", marker = ".")

plt.legend(loc = 1)
plt.xticks(rotation = 90)
plt.tight_layout()
plt.show()

Example

Example mock evaluation is shown below (examples/example_benchmark.py): Example

SkBlaz / san

readme

Feature ranking with self-attention networks

Installing SANs

Using SANs