MNIST/Pylearn2 Simple 2-layered MNIST runner

gistfile1.txt

import struct
import numpy as np
from pylearn2.utils import serial
import theano


def read_mnist_images():
    path = "mnist/"
    label_path = path + 't10k-images-idx3-ubyte'
    dtype = 'float32'
    with open(label_path, 'rb') as f:
        magic, number, rows, cols = struct.unpack('>iiii', f.read(16))
        array = np.fromfile(f, dtype='uint8').reshape((number, rows, cols))
    dtype = np.dtype(dtype)
    array = array.astype(dtype)
    return array


def get_testing_input():
    axes = ['b', 0, 1, 'c']
    topo_view = read_mnist_images()
    m, r, c = topo_view.shape
    topo_view = topo_view.reshape(m, r, c, 1)
    default = ('b', 0, 1, 'c')
    return topo_view.transpose(*[default.index(axis) for axis in axes])


def get_testing_labels():
    path = "mnist/"
    label_path = path + 't10k-labels-idx1-ubyte'
    with open(label_path, 'rb') as f:
        _, number = struct.unpack('>ii',
                                  f.read(8))  # gotta do this line first to move the point to AFTER the first 8 bytes
        label_array = np.fromfile(f, dtype='uint8')
        label_array = np.atleast_2d(label_array).transpose()
    return label_array


if __name__ == "__main__":
    #load model from file
    neural_network_model = serial.load("./model.saved")

    # test trained model
    inputs = get_testing_input()
    targets = get_testing_labels()
    sample_input = [inputs[0].ravel()] #flatten the array

    #make it into an input that the model can accept
    sample_input = np.array(sample_input)
    dtype = np.dtype("float32")
    sample_input = sample_input.astype(dtype)

    print neural_network_model.fprop(theano.shared(sample_input)).eval() #the first input should resovle to 7.