decision_boundary_plot.py

# Print the decision boundary of a neural network.

import numpy as np
import matplotlib.pyplot as plt
from matplotlib.colors import ListedColormap


def one_hot_encode(Y):
    n_labels = Y.shape[0]
    result = np.zeros((n_labels, 2))
    for i in range(n_labels):
        result[i][Y[i]] = 1
    return result


# Uncomment one of the next three lines to decide which dataset to load
x1, x2, y = np.loadtxt('linearly_separable.txt', skiprows=1, unpack=True)

# not a straight line, the accuracy is no longer 100%
# x1, x2, y = np.loadtxt('non_linearly_separable.txt', skiprows=1, unpack=True)

# x1, x2, y = np.loadtxt('circles.txt', skiprows=1, unpack=True)

X_train = X_test = np.column_stack((x1, x2))
Y_train_unencoded = Y_test = y.astype(int).reshape(-1, 1)
Y_train = one_hot_encode(Y_train_unencoded)


# Generate a mesh over one-dimensional data
# (The mesh() and plot_boundary() functionality were inspired by the
# documentation of the BSD-licensed scikit-learn library.)
def mesh(values):
    range = values.max() - values.min()
    padding_percent = 5
    padding = range * padding_percent * 0.01
    resolution = 1000
    interval = (range + 2 * range * padding) / resolution
    return np.arange(values.min() - padding, values.max() + padding, interval)


def plot_data_by_label(input_variables, labels, label_selector, symbol):
    points = input_variables[(labels == label_selector).flatten()]
    plt.plot(points[:, 0], points[:, 1], symbol, markersize=4)


plot_data_by_label(X_train, Y_train_unencoded, 0, 'bs')
plot_data_by_label(X_train, Y_train_unencoded, 1, 'g^')
plt.gca().axes.set_xlabel("Input A", fontsize=20)
plt.gca().axes.set_ylabel("Input B", fontsize=20)
plt.gca().axes.xaxis.set_ticklabels([])
plt.gca().axes.yaxis.set_ticklabels([])
plt.show()