pca.py

from utils import *


class PCA:
    '''
    A general PCA class
    '''
    proj_matrix = None

    def __init__(self, X, explain_ratio=0.95):
        cov_matrix = np.cov(X.T)
        eigen_values, eigen_vectors = np.linalg.eig(cov_matrix)
        esum = np.sum(eigen_values)
        variance_explained = np.zeros(eigen_values.shape[0])
        for i, v in enumerate(eigen_values):
            variance_explained[i] = v / sum(eigen_values)
        cumulative_variance_explained = np.cumsum(variance_explained)
        self.proj_matrix = eigen_vectors[:, :max(1, np.where(explain_ratio <= cumulative_variance_explained)[0][0])]
        print("Dimension reduced from %d to %d" % (len(eigen_values), self.proj_matrix.shape[1]))
        return

    def proj(self, x):
        return np.real(np.dot(x, self.proj_matrix))


if __name__ == "__main__":
    # Test
    X = np.random.random((5, 10))
    print(X)
    pca = PCA(X, 0.85)
    X_pca = pca.proj(X)
    print(X_pca)