The objective of this assignment is to learn about simple data curation practices, and familiarize you with some of the data we'll be reusing later.
This notebook uses the notMNIST dataset to be used with python experiments. This dataset is designed to look like the classic MNIST dataset, while looking a little more like real data: it's a harder task, and the data is a lot less 'clean' than MNIST.
In [2]:
# These are all the modules we'll be using later. Make sure you can import them
# before proceeding further.
from __future__ import print_function
import matplotlib.pyplot as plt
import numpy as np
import os
import sys
import tarfile
from IPython.display import display, Image
from scipy import ndimage
from sklearn.linear_model import LogisticRegression
from six.moves.urllib.request import urlretrieve
from six.moves import cPickle as pickle
import random # show random images
import glob # Problem 2
# Config the matplotlib backend as plotting inline in IPython
%matplotlib inline
First, we'll download the dataset to our local machine. The data consists of characters rendered in a variety of fonts on a 28x28 image. The labels are limited to 'A' through 'J' (10 classes). The training set has about 500k and the testset 19000 labelled examples. Given these sizes, it should be possible to train models quickly on any machine.
In [5]:
# course-harcoded url
url = 'http://commondatastorage.googleapis.com/books1000/'
last_percent_reported = None
def download_progress_hook(count, blockSize, totalSize):
"""A hook to report the progress of a download. This is mostly intended for users with
slow internet connections. Reports every 5% change in download progress.
"""
global last_percent_reported
percent = int(count * blockSize * 100 / totalSize)
if last_percent_reported != percent:
if percent % 5 == 0:
sys.stdout.write("%s%%" % percent)
sys.stdout.flush()
else:
sys.stdout.write(".")
sys.stdout.flush()
last_percent_reported = percent
def maybe_download(filename, expected_bytes, force=False):
"""Download a file if not present, and make sure it's the right size."""
# Create path where the data file will be stored
# get working directory and go to the parent - distro agnostic code
dpath = os.path.abspath(os.path.join(os.getcwd(), os.pardir))
# go to data directory
dpath = os.path.join(dpath, 'data')
# get filepath
fpath = os.path.join(dpath, filename)
# Download file if needed
if force or not os.path.exists(fpath):
print('Attempting to download:', filename)
filename, _ = urlretrieve(url + filename, filename, reporthook=download_progress_hook)
print('\nDownload Complete!')
# move new file from working directory to data location
# current file location
cpath = os.path.join(os.getcwd(), filename)
os.rename(cpath, fpath)
# check existing file if it exists or new if not
statinfo = os.stat(fpath)
# Verify file size.
if statinfo.st_size == expected_bytes:
print('Found and verified', filename)
else:
raise Exception(
'Failed to verify ' + filename +
'. Can you get to it with a browser or download it again?')
return (filename)
train_filename = maybe_download('notMNIST_large.tar.gz', 247336696)
test_filename = maybe_download('notMNIST_small.tar.gz', 8458043)
Extract the dataset from the compressed .tar.gz file. This should give you a set of directories, labelled A through J.
In [6]:
num_classes = 10
np.random.seed(133)
def maybe_extract(filename, force=False):
# get new dir name
dirn = os.path.splitext(os.path.splitext(filename)[0])[0] # remove .tar.gz
# Create data directory path
dpath = os.path.abspath(os.path.join(os.getcwd(), os.pardir))
dpath = os.path.join(dpath, 'data')
# Create dir to unzip data
dirn = os.path.join(dpath, dirn)
# Create zipped file path
fpath = os.path.join(dpath, filename)
if os.path.isdir(dirn) and not force:
# You may override by setting force=True.
print('%s already present - Skipping extraction of %s.' % (dirn, filename))
else:
print('Extracting data for %s. This may take a while. Please wait.' % dirn)
tar = tarfile.open(fpath)
sys.stdout.flush()
# set path so data are extracted within the data folder.
tar.extractall(path=dpath)
tar.close()
data_folders = [
os.path.join(dirn, d) for d in sorted(os.listdir(dirn))
if os.path.isdir(os.path.join(dirn, d))]
if len(data_folders) != num_classes:
raise Exception(
'Expected %d folders, one per class. Found %d instead.' % (
num_classes, len(data_folders)))
print("Data folders list:")
print(data_folders)
return(data_folders)
train_folders = maybe_extract(train_filename)
test_folders = maybe_extract(test_filename)
In [7]:
def show_random_images(numi):
"""
Function to display a specified number of random images from the extracted dataset.
Arguments:
numi: Integer, how many images to show.
"""
# First let's create a list of all the files.
# Create data directory path
dpath = os.path.abspath(os.path.join(os.getcwd(), os.pardir))
dpath = os.path.join(dpath, 'data')
# notMNIST_small directory:
dsmall = os.path.join(dpath, 'notMNIST_small')
# notMNIST_large directory:
dlarge = os.path.join(dpath, 'notMNIST_large')
# create $numi random number of paths of images
name1 = []
it1 = 0
while it1 < numi:
# select random notMNIST
rpath0 = random.choice([dlarge, dsmall])
# select random letter
rpath1 = random.choice(["A/", "B/", "C/", "D/", "E/", "F/", "H/", "I/", "J/"])
# join them
rpath = os.path.join(rpath0, rpath1)
# select random image from files
onlyfiles = [fi for fi in os.listdir(rpath) if os.path.isfile(os.path.join(rpath, fi))]
name2 = random.choice(onlyfiles)
# add that random name to it's path
name2 = os.path.join(rpath, name2)
# add it to list of images
name1.append(name2)
it1 += 1
for it2 in name1:
print("Showing Image from path:\n" + it2)
im1 = Image(filename=(it2))
display(im1)
# show me 10 images
show_random_images(10)
Now let's load the data in a more manageable format. Since, depending on your computer setup you might not be able to fit it all in memory, we'll load each class into a separate dataset, store them on disk and curate them independently. Later we'll merge them into a single dataset of manageable size.
We'll convert the entire dataset into a 3D array (image index, x, y) of floating point values, normalized to have approximately zero mean and standard deviation ~0.5 to make training easier down the road.
A few images might not be readable, we'll just skip them.
In [8]:
### Image preprocessing happening in this step !!! ###
image_size = 28 # Pixel width and height.
pixel_depth = 255.0 # Number of levels per pixel.
def load_letter(folder, min_num_images):
"""Load the data for a single letter label."""
image_files = os.listdir(folder)
dataset = np.ndarray(shape=(len(image_files), image_size, image_size),
dtype=np.float32)
print(folder)
num_images = 0
for image in image_files:
image_file = os.path.join(folder, image)
try:
# read image as array:
# https://docs.scipy.org/doc/scipy-0.18.1/reference/generated/scipy.ndimage.imread.html
# code below also shifts average to 0 and standard deviation to 1
# This scaling happens assuming the 255 pixel depth is uniformly populated.
image_data = (ndimage.imread(image_file).astype(float) -
pixel_depth / 2) / pixel_depth
if image_data.shape != (image_size, image_size):
raise Exception('Unexpected image shape: %s' % str(image_data.shape))
dataset[num_images, :, :] = image_data
num_images = num_images + 1
except IOError as e:
print('Could not read:', image_file, ':', e, '- it\'s ok, skipping.')
dataset = dataset[0:num_images, :, :]
if num_images < min_num_images:
raise Exception('Many fewer images than expected: %d < %d' %
(num_images, min_num_images))
print('Full dataset tensor:', dataset.shape)
print('Mean:', np.mean(dataset))
print('Standard deviation:', np.std(dataset))
return(dataset)
def maybe_pickle(data_folders, min_num_images_per_class, force=False):
dataset_names = []
for folder in data_folders:
set_filename = folder + '.pickle'
dataset_names.append(set_filename)
if os.path.exists(set_filename) and not force:
# You may override by setting force=True.
print('%s already present - Skipping pickling.' % set_filename)
else:
print('Pickling %s.' % set_filename)
dataset = load_letter(folder, min_num_images_per_class)
try:
with open(set_filename, 'wb') as f:
pickle.dump(dataset, f, pickle.HIGHEST_PROTOCOL)
except Exception as e:
print('Unable to save data to', set_filename, ':', e)
return(dataset_names)
train_datasets = maybe_pickle(train_folders, 50000)
test_datasets = maybe_pickle(test_folders, 1800)
In [9]:
## Open a random image from the pickled files.
def show_rnd_pkl_image():
"""Function that shows a random pickled image."""
# First let's create a list of all the files.
# Create data directory path
dpath = os.path.abspath(os.path.join(os.getcwd(), os.pardir))
dpath = os.path.join(dpath, 'data')
# notMNIST_small directory:
dsmall = os.path.join(dpath, 'notMNIST_small')
# notMNIST_large directory:
dlarge = os.path.join(dpath, 'notMNIST_large')
# Find all pickle files in each directory.
# http://stackoverflow.com/a/3215392
# Create a list of all nonMNIST_small pickles
lsmall = glob.glob(dsmall + '/*.pickle')
# Create a list of all nonMNIST_large pickles
llarge = glob.glob(dlarge + '/*.pickle')
# Pick a random pickle to load (either large or small !)
rpklfile = random.choice([lsmall, llarge])
rpklfile = random.choice(rpklfile)
# verify randomness
print(rpklfile)
with open(rpklfile, 'rb') as rf:
imgPkl = pickle.load(rf)
plt.imshow(random.choice(list(imgPkl)))
show_rnd_pkl_image()
In [10]:
def disp_number_images(data_folders):
for folder in data_folders:
pickle_filename = folder + '.pickle'
try:
with open(pickle_filename, 'rb') as f:
dataset = pickle.load(f)
except Exception as e:
print('Unable to read data from', pickle_filename, ':', e)
return
print('Number of images in ', folder, ' : ', len(dataset))
disp_number_images(train_folders)
disp_number_images(test_folders)
Merge and prune the training data as needed. Depending on your computer setup, you might not be able to fit it all in memory, and you can tune train_size as needed. The labels will be stored into a separate array of integers 0 through 9.
Also create a validation dataset for hyperparameter tuning.
In [11]:
def make_arrays(nb_rows, img_size):
if nb_rows:
dataset = np.ndarray((nb_rows, img_size, img_size), dtype=np.float32)
labels = np.ndarray(nb_rows, dtype=np.int32)
else:
dataset, labels = None, None
return dataset, labels
def merge_datasets(pickle_files, train_size, valid_size=0):
num_classes = len(pickle_files)
valid_dataset, valid_labels = make_arrays(valid_size, image_size)
train_dataset, train_labels = make_arrays(train_size, image_size)
vsize_per_class = valid_size // num_classes
tsize_per_class = train_size // num_classes
start_v, start_t = 0, 0
end_v, end_t = vsize_per_class, tsize_per_class
end_l = vsize_per_class+tsize_per_class
for label, pickle_file in enumerate(pickle_files):
try:
with open(pickle_file, 'rb') as f:
letter_set = pickle.load(f)
# let's shuffle the letters to have random validation and training set
np.random.shuffle(letter_set)
if valid_dataset is not None:
valid_letter = letter_set[:vsize_per_class, :, :]
valid_dataset[start_v:end_v, :, :] = valid_letter
valid_labels[start_v:end_v] = label
start_v += vsize_per_class
end_v += vsize_per_class
train_letter = letter_set[vsize_per_class:end_l, :, :]
train_dataset[start_t:end_t, :, :] = train_letter
train_labels[start_t:end_t] = label
start_t += tsize_per_class
end_t += tsize_per_class
except Exception as e:
print('Unable to process data from', pickle_file, ':', e)
raise
return valid_dataset, valid_labels, train_dataset, train_labels
train_size = 500000
valid_size = 29000
test_size = 18000
valid_dataset, valid_labels, train_dataset, train_labels = merge_datasets(
train_datasets, train_size, valid_size)
_, _, test_dataset, test_labels = merge_datasets(test_datasets, test_size)
print('Training:', train_dataset.shape, train_labels.shape)
print('Validation:', valid_dataset.shape, valid_labels.shape)
print('Testing:', test_dataset.shape, test_labels.shape)
Next, we'll randomize the data. It's important to have the labels well shuffled for the training and test distributions to match.
In [12]:
def randomize(dataset, labels):
permutation = np.random.permutation(labels.shape[0])
shuffled_dataset = dataset[permutation,:,:]
shuffled_labels = labels[permutation]
return (shuffled_dataset, shuffled_labels)
train_dataset, train_labels = randomize(train_dataset, train_labels)
test_dataset, test_labels = randomize(test_dataset, test_labels)
valid_dataset, valid_labels = randomize(valid_dataset, valid_labels)
To be sure that the data are still fine after the merger and the randomization, we will select one item and display the image alongside the label.
Note: 0 = A, 1 = B, 2 = C, 3 = D, 4 = E, 5 = F, 6 = G, 7 = H, 8 = I, 9 = J.
In [13]:
pretty_labels = {0: 'A', 1: 'B', 2: 'C', 3: 'D', 4: 'E', 5: 'F', 6: 'G', 7: 'H', 8: 'I', 9: 'J'}
def disp_sample_dataset(dataset, labels):
items = random.sample(range(len(labels)), 8)
for i, item in enumerate(items):
plt.subplot(2, 4, i+1)
plt.axis('off')
plt.title(pretty_labels[labels[item]])
plt.imshow(dataset[item])
In [14]:
disp_sample_dataset(train_dataset, train_labels)
In [15]:
disp_sample_dataset(train_dataset, train_labels)
Finally, let's save the data for later reuse:
In [19]:
# Create data directory path
dpath = os.path.abspath(os.path.join(os.getcwd(), os.pardir))
dpath = os.path.join(dpath, 'data')
# create pickle data file path
pickle_file = os.path.join(dpath,'notMNIST.pickle')
# save data if they aren't already saved or forced.
def maybe_save_data(filepath, force=False):
# Download file if needed
if force or not os.path.exists(filepath):
print('Attempting to save data at:\n', filepath)
try:
f = open(pickle_file, 'wb')
save = {
'train_dataset': train_dataset,
'train_labels': train_labels,
'valid_dataset': valid_dataset,
'valid_labels': valid_labels,
'test_dataset': test_dataset,
'test_labels': test_labels,
}
pickle.dump(save, f, pickle.HIGHEST_PROTOCOL)
f.close()
print('Data saved.')
except Exception as e:
print('Unable to save data to', pickle_file, ':', e)
raise
else:
print('Data has been processed and saved in previous run.')
# Note: Previous run reshuffling will likely be different
# from current run!
statinfo = os.stat(pickle_file)
print('Compressed pickle size:', statinfo.st_size)
maybe_save_data(pickle_file)
By construction, this dataset might contain a lot of overlapping samples, including training data that's also contained in the validation and test set! Overlap between training and test can skew the results if you expect to use your model in an environment where there is never an overlap, but are actually ok if you expect to see training samples recur when you use it. Measure how much overlap there is between training, validation and test samples.
Optional questions:
In this part, we will explore the datasets and understand better the overlap cases. There are overlaps, but there are also duplicates in the same dataset! Processing time is also critical. We will first use nested loops and matrix comparison, which is slow.
Potential enhancement: Use a hash function to accelerate and process the whole dataset.
In [20]:
def display_overlap(overlap, source_dataset, target_dataset):
item = random.choice(list(overlap.keys()))
imgs = np.concatenate(([source_dataset[item]], target_dataset[overlap[item][0:7]]))
plt.suptitle(item)
for i, img in enumerate(imgs):
plt.subplot(2, 4, i+1)
plt.axis('off')
plt.imshow(img)
def extract_overlap(dataset_1, dataset_2):
overlap = {}
for i, img_1 in enumerate(dataset_1):
for j, img_2 in enumerate(dataset_2):
if np.array_equal(img_1, img_2):
if not i in overlap.keys():
overlap[i] = []
overlap[i].append(j)
return overlap
In [21]:
%time overlap_test_train = extract_overlap(test_dataset[:200], train_dataset)
In [22]:
print('Number of overlaps:', len(overlap_test_train.keys()))
display_overlap(overlap_test_train, test_dataset[:200], train_dataset)
Let's get an idea of what an off-the-shelf classifier can give you on this data. It's always good to check that there is something to learn, and that it's a problem that is not so trivial that a canned solution solves it.
Train a simple model on this data using 50, 100, 1000 and 5000 training samples. Hint: you can use the LogisticRegression model from sklearn.linear_model.
Optional question: train an off-the-shelf model on all the data!
In [23]:
def tryLogRegr(sample_size):
"""
Arguments:
sample_size: Integer to determine sample size
"""
regr = LogisticRegression()
X_test = test_dataset.reshape(test_dataset.shape[0], 28 * 28)
y_test = test_labels
X_train = train_dataset[:sample_size].reshape(sample_size, 784)
y_train = train_labels[:sample_size]
%time regr.fit(X_train, y_train)
rscore = regr.score(X_test, y_test)
print("Mean acccuracy of the linear regression model is: {}"
.format(rscore))
pred_labels = regr.predict(X_test)
disp_sample_dataset(test_dataset, pred_labels)
tryLogRegr(50)
In [24]:
tryLogRegr(100)
In [25]:
tryLogRegr(1000)
In [26]:
tryLogRegr(5000)
Looks like ~85% is the limit for a linear method
Let's try on all of the dataset.
In [27]:
def tryLogRegrAll():
"""
Function to perform Logistic Regression on all our dataset.
"""
# sag solver works better for bigger datasets
# n_jobs = -2 automatically selects (max - 1) available cores!
# using -1
regr = LogisticRegression(solver='sag', n_jobs = -1)
X_test = test_dataset.reshape(test_dataset.shape[0], 28 * 28)
y_test = test_labels
X_train = train_dataset.reshape(train_dataset.shape[0], 784)
y_train = train_labels
%time regr.fit(X_train, y_train)
rscore = regr.score(X_test, y_test)
print("Mean acccuracy of the linear regression model is: {}"
.format(rscore))
pred_labels = regr.predict(X_test)
disp_sample_dataset(test_dataset, pred_labels)
tryLogRegrAll()