Comparison of LDA transforms

example_lda.py

import numpy as np
from sklearn.lda import LDA
from sklearn.datasets import make_blobs
from sklearn.utils import check_array
from sklearn.svm import LinearSVC
import matplotlib.pyplot as plt


def new_transform(lda, X):
    X = check_array(X)
    # center and scale data
    X_new = np.dot(X - lda.xbar_, lda.scalings_)
    n_components = X.shape[1] if lda.n_components is None \
        else lda.n_components
    return X_new[:, :n_components]


def old_transform(lda, X):
    X = check_array(X)
    # center and scale data
    X = np.dot(X - lda.xbar_, lda.scalings_)
    n_comp = X.shape[1] if lda.n_components is None \
        else lda.n_components
    return np.dot(X, lda.coef_[:n_comp].T)


def plot_transform(X, y, transform, title):
    lda = LDA(n_components=2).fit(X, y)
    Z = transform(lda, X)
    plt.scatter(Z[:, 0], Z[:, 1], c=y)
    plt.gca().set_aspect('equal')
    acc = LinearSVC().fit(Z, y).score(Z, y)
    plt.title(title + ' LinearSVC score: ' + str(acc))

X, y = make_blobs(1000, centers=3, center_box=(-2, 2), random_state=1999)

plt.subplot(2, 2, 1)
plot_transform(X, y, old_transform, 'old')

plt.subplot(2, 2, 3)
plot_transform(X, y, new_transform, 'new')

X, y = make_blobs(1000, centers=3, center_box=(-10, 10), random_state=1999)

plt.subplot(2, 2, 2)
plot_transform(X, y, old_transform, 'old')

plt.subplot(2, 2, 4)
plot_transform(X, y, new_transform, 'new')

plt.show()