decision-boundary-mapper 0.2.3

A tool for visualizing the decision boundary of a machine learning model.

This is the package provides functionality for visualizing the classifiers decision boundaries.

It is based on the work of Cristian Grosu for the master thesis project for 2023 at Utrecht University. If you use this package, please cite the following paper: [placeholder for the paper]

The package is available on PyPI and can be installed using pip: pip install decision-boundary-mapper

Usage exmaples:

1. This package comes with a simple GUI that allows you to visualize the decision boundaries of a classifier. The GUI is based on the PySimpleGUI package and can be started by running the following code:

from decision_boundary_mapper import GUI

GUI().start()

2. The package comes with two examples of complete pipelines for visualizing the decision boundaries of a classifier. Both examples use MNIST (handwritten digits) dataset. The first example DBM_usage_example uses t-SNE to project the data from the nD space to the 2D space, then neural network is trained to fit the inverse projection from 2D to nD and the decision boundaries are visualized using the 2D projection. The second example DBM_usage_example_2 uses UMAP to project the data from the nD space to the 2D space, then neural network is trained to fit the inverse projection from 2D to nD and the decision boundaries are visualized using the 2D projection. After which a simple classifier is used to color each point of the 2D projection. The second example SDM_usage_example uses an Autoencoder Neural Network to project the data from the nD space to the 2D space, then a simple classifier is used to color each point of the 2D projection. The examples can be found in the examples folder.

from decision_boundary_mapper import DBM_usage_example, SDM_usage_example

DBM_usage_example() # run the first example
SDM_usage_example() # run the second example

3. The package main functionality comes in two classes DBM (i.e. learns inverse projection when a 2D projection is given) and SDBM (i.e. learns both the projection and the inverse projection). The classes can be used as follows:

from decision_boundary_mapper import DBM, SDBM
from matplotlib import pyplot as plt

# load the data
...
X_train, X_test, y_train, y_test = load_data() 
...
# create a simple neural network
...
classifier = ... # for compatibility with the package the classifier should be constructed using tensorflow.keras
...

dbm = DBM(classifier) # create a DBM object
img, img_confidence, _, _ = dbm.generate_decision_boundary(X_train, y_train, X_test, y_test, resolution = 256) # generate the decision boundary

sdbm = SDBM(classifier) # create a SDBM object
img, img_confidence, _, _ = sdbm.generate_decision_boundary(X_train, y_train, X_test, y_test, resolution = 256) # generate the decision boundary
...
# visualize the decision boundaries
plt.imshow(img)
plt.show()

Created by Cristian Grosu for the master thesis project for 2023 at Utrecht University