import numpy as np
import matplotlib.pyplot as plt
from keras import backend as K
from random import shuffle
from pnslib import utils
from pnslib import ml
# Load T-shirt/top and Trouser classes from Fashion MNIST
# complete label description is at
# https://github.com/zalandoresearch/fashion-mnist#labels
(train_x, train_y, test_x,
test_y) = utils.binary_fashion_mnist_load(class_list=[0, 1], flatten=True)
print("[MESSAGE] Dataset is loaded.")
# preprocessing for training and testing images
train_x = train_x.astype("float32") / 255. # rescale image
mean_train_x = np.mean(train_x, axis=0) # compute the mean across pixels
train_x -= mean_train_x # remove the mean pixel value from image
test_x = test_x.astype("float32") / 255.
test_x -= mean_train_x
print("[MESSAGE] Dataset is preporcessed.")
# Use PCA to reduce the dimension of the dataset,
# so that the training will be less expensive
# perform PCA on training dataset
"""Demonstrate the usage of Fashion-MNIST. Author: Yuhuang Hu Email : [email protected] """ from __future__ import print_function, absolute_import import matplotlib.pyplot as plt import pnslib.utils as utils (train_x, train_y, test_x, test_y) = utils.binary_fashion_mnist_load() print(train_x.shape) print(train_y.shape) print(test_x.shape) print(test_y.shape) plt.figure() for idx in range(100): plt.imshow(test_x[idx, ..., 0], cmap="gray") plt.show()
"""Perform PCA on Fashion-MNIST. Author: Yuhuang Hu Email : [email protected] """ from __future__ import print_function, absolute_import import numpy as np from pnslib import utils from pnslib import ml # load data (train_x, train_y, test_x, test_y) = utils.binary_fashion_mnist_load(flatten=True) train_x = train_x.astype("float32") / 255. mean_train_x = np.mean(train_x, axis=0) train_x -= mean_train_x test_x = train_x.astype("float32") / 255. test_x -= mean_train_x print(train_x.dtype) print(test_x.dtype) # perform PCA on training dataset train_x, R, n_retained = ml.pca(train_x) # perform PCA on testing dataset test_x = ml.pca_fit(test_x, R, n_retained)