digits loader

load_digits.py

# Data from : http://www.metz.supelec.fr//metz/personnel/frezza/ApprentissageNumerique/TP-MachineLearning/dig_app_text.cb
# Data are said to be provided UCI Machine Learning Repository ... but I did not find the dataset on the repository.. 
import numpy as np

digit_idx = 0
header = 5
width = 28
height = 28
num_digits = 4000
filename = 'dig_app_text.cb'

samples = np.zeros(num_digits, dtype=[('input',  float, width*height), ('label', int, 1)])
with open(filename) as digits:
    line = digits.readline()
    for i in range(header):
        line = digits.readline()
    while line:
        # Parse the digit
        digit = []
        for i in range(height):
            digit += map(int, line.split())
            line = digits.readline()
        samples['input'][digit_idx, :] = np.array(digit) / 255.

        # Parse the label
        label = list(map(int, line.split())).index(1)
        samples['label'][digit_idx] = label

        digit_idx += 1
        line = digits.readline()

np.save('dig_app_text.cb.npy', samples)

import matplotlib.pyplot as plt
fig, axes = plt.subplots(1,10, figsize=(10,2))
for i in range(10):
    axes[i].imshow(samples['input'][i].reshape((width, height)) , cmap='gray_r')
    axes[i].set_xticks([])
    axes[i].set_yticks([])
    axes[i].set_title('I\'m a {}'.format(samples['label'][i]))
plt.savefig('digits.png', bbox_inches='tight')
plt.show()

Get the data from dig_app_text.cb and copy it close to the script.

python3 load_digits.py