add helper scripts to use LeNet model against MNIST data

[lelayf]

This is taking over PR #1017 to target BVLC:dev branch instead of BVLC:master.

[sergeyk]

This looks good to me.

[shelhamer]

@lelayf thanks for the helper scripts to improve the LeNet example. Please take a look at #959 and the IPython notebook there. It has simple code for loading the MNIST digits and labels that you could integrate here.

TEST_DATA_FILE = '../../data/mnist/t10k-images-idx3-ubyte'
TEST_LABEL_FILE = '../../data/mnist/t10k-labels-idx1-ubyte'
n = 10000

with open(TEST_DATA_FILE, 'rb') as f:
    f.read(16) # skip the header
    images = np.fromstring(f.read(n * 28*28), dtype=np.uint8)
images = images.reshape(n, 28*28)

with open(TEST_LABEL_FILE, 'rb') as f:
    f.read(8) # skip the header
    labels = np.fromstring(f.read(n), dtype=np.uint8)

# scale and reshape
images = images.reshape(n, 1, 28, 28).astype(np.float32) / 255.   

[lelayf]

@shelhamer I checked the IPython notebook you mentioned and changed the way I read those bytes to something hopefully more expressive. Depending on exceptions that are raised at runtime, the reader might return the expected arrays or nothing. For the current purpose it seems fine but if you'd rather see less indentation of the return statement to return None,None in case of failure then I can make the change.

[shelhamer]

Thanks for the helper scripts, and for switching the base of your PR for our workflow at the time, but in the end dev was deprecated for a single branch workflow so I'm closing this PR.

Improvements to our examples and documentation are of course always welcome in master.