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 [1]:
# 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
# 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 [2]:
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."""
if force or not os.path.exists(filename):
print('Attempting to download:', filename)
filename, _ = urlretrieve(url + filename, filename, reporthook=download_progress_hook)
print('\nDownload Complete!')
statinfo = os.stat(filename)
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?')
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 [2]:
num_classes = 10
np.random.seed(133)
def maybe_extract(filename, force=False):
root = os.path.splitext(os.path.splitext(filename)[0])[0] # remove .tar.gz
if os.path.isdir(root) and not force:
# You may override by setting force=True.
print('%s already present - Skipping extraction of %s.' % (root, filename))
else:
print('Extracting data for %s. This may take a while. Please wait.' % root)
tar = tarfile.open(filename)
sys.stdout.flush()
tar.extractall()
tar.close()
data_folders = [
os.path.join(root, d) for d in sorted(os.listdir(root))
if os.path.isdir(os.path.join(root, 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)
return data_folders
train_folders = maybe_extract(train_filename)
test_folders = maybe_extract(test_filename)
In [19]:
import os, sys
print( os.getcwd() ) # let's see where we are in the local directory
os.listdir( os.getcwd() )
Out[19]:
In [10]:
notMNIST_smallAlistfiles = os.listdir('./notMNIST_small/A'); len( notMNIST_smallAlistfiles) # lots of files!
Out[10]:
In [14]:
notMNIST_smallAlistfiles.index("Q0NXaWxkV29yZHMtQm9sZEl0YWxpYy50dGY=.png") != 0 # check to see if it's there or not
Out[14]:
How to Include image or picture in jupyter notebook
In [22]:
Huh? Let's try iPython
In [23]:
from IPython.display import display, Image
In [25]:
display(Image(filename="notMNIST_small/A/Q0NXaWxkV29yZHMtQm9sZEl0YWxpYy50dGY=.png"))
In [26]:
display(Image(filename="notMNIST_small/B/Q2FsaWd1bGEgUmVndWxhci50dGY=.png"))
In [27]:
display(Image(filename="notMNIST_small/C/QmVlc2tuZWVzQy5vdGY=.png"))
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 [3]:
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:
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, 45000)
test_datasets = maybe_pickle(test_folders, 1800)
cf. Problem 2: Verify normalized images
EY : 20170126 Notice this line in load_letter above:
image_data = (ndimage.imread(image_file).astype(float) - pixel_depth / 2) / pixel_depthas noted by Vincent in the Instructor's comments below the video.
We've then stacked all the images for each letter into a giant three-dimensional tensor and pickled it. Saving these tensors to disk allows us to reduce our memory footprint.
dataset[num_images, :, :] = image_data
num_images = num_images + 1
...
with open(set_filename, 'wb') as f:
pickle.dump(dataset, f, pickle.HIGHEST_PROTOCOL)Now, to retrieve a normalized image, we need to read back a pickle file (there is one per dataset folder or letter) into a tensor and grab a slice from it. We can then display the slice using imshow() from matplotlib.pyplot (imported previously as plt).
In [28]:
pickle_file = train_datasets[0] # index 0 should be all As, 1 = all Bs, etc.
In [29]:
with open(pickle_file, 'rb') as f:
letter_set = pickle.load(f) # unpickle
sample_idx = np.random.randint(len(letter_set)) # pick a random image index
sample_image = letter_set[sample_idx, :, :] # extract a 2D slice
plt.figure()
plt.imshow(sample_image) # display it
In [35]:
print( len(letter_set)); # length of letter_set
plt.imshow( letter_set[0,:,:]); plt.imshow( letter_set[ len(letter_set)/26*18,:,:])
Out[35]:
In [37]:
f.close()
In [49]:
pickle_files = [train_datasets[i] for i in range(len(train_datasets))] # A,B,C,D,E,F,G,H,I,J
In [50]:
with open(pickle_files[8], 'rb') as f:
letter_set = pickle.load(f) # unpickle
sample_idx = np.random.randint(len(letter_set)) # pick a random image index
sample_image = letter_set[sample_idx, :, :] # extract a 2D slice
plt.figure()
plt.imshow(sample_image) # display it
f.close()
In [42]:
len(train_datasets)
Out[42]:
In [52]:
for i in range(len(train_datasets)):
print("For training set", i)
with open( train_datasets[i], 'rb') as f:
letter_set = pickle.load(f) # unpickle
print( " , ", len(letter_set), " \n")
f.close()
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 [53]:
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 = 200000
valid_size = 10000
test_size = 10000
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 [54]:
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)
In [61]:
print( type(train_dataset)); print( train_dataset.shape ); print( train_dataset.dtype);
print( type(train_labels)); print( train_labels.dtype )
In [73]:
print(np.mean(train_dataset, axis=0).shape); print(np.mean(train_dataset).shape); np.mean(train_dataset)
Out[73]:
In [74]:
print(np.mean(test_dataset, axis=0).shape); print(np.mean(test_dataset).shape); np.mean(test_dataset)
Out[74]:
In [76]:
print( np.mean(valid_dataset) ) ; print( np.var(train_dataset) ) ; print( np.var(test_dataset) ) ;
print( np.var(valid_dataset) )
In [79]:
histtrain, binstrain = np.histogram( train_dataset)
In [83]:
plt.xlim(min(binstrain), max(binstrain))
plt.bar( binstrain[:-1], histtrain, width = 1)
#plt.hist( histtrain, bins=binstrain)
#plt.bar(bin_edges[:-1], hist, width = 1)
Out[83]:
In [87]:
histtest, binstest = np.histogram( test_dataset)
plt.bar( binstest[:-1], histtest )
Out[87]:
In [88]:
histvalid, binsvalid = np.histogram( valid_dataset)
plt.bar( binsvalid[:-1], histvalid )
Out[88]:
In [92]:
print( histvalid )
binsvalid
Out[92]:
Finally, let's save the data for later reuse:
In [55]:
pickle_file = 'notMNIST.pickle'
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()
except Exception as e:
print('Unable to save data to', pickle_file, ':', e)
raise
In [56]:
statinfo = os.stat(pickle_file)
print('Compressed pickle size:', statinfo.st_size)
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 [93]:
train_dataset.flags
Out[93]:
In [108]:
train_dataset.flags.writeable=False
test_dataset.flags.writeable=False
In [111]:
dup_table = {}
In [112]:
for idx, img in enumerate(train_dataset):
h = hash(img.data)
# if h in dup_table and (train_dataset[dup_table[h]].data == img.data):
# print( "Duplicate image: %d matches %d" % (idx, dup_table[h]) )
# print (idx, dup_table[h])
dup_table[h] = idx
In [113]:
howmanyduplicatesoftestintrain = 0
for idx, img in enumerate(test_dataset):
h = hash(img.data)
if h in dup_table and (train_dataset[dup_table[h]].data == img.data):
# print( "Test image %d is in the training set" % idx)
howmanyduplicatesoftestintrain+=1
print(howmanyduplicatesoftestintrain)
In [115]:
help(hash)
In [122]:
import time
def fast_overlaps_num_set_and_hash(images1,images2):
images1.flags.writeable=False
images2.flags.writeable=False
hash1 = set([hash(image1.data) for image1 in images1])
hash2 = set([hash(image2.data) for image2 in images2])
all_overlaps = set.intersection(hash1, hash2)
return len(all_overlaps)
def find_dups_and_overlaps(images1, images2):
images1.flags.writeable=False
images2.flags.writeable=False
dup_table={}
duplicates1 = []
for idx, img in enumerate(images1):
h = hash(img.data)
if h in dup_table and (images1[dup_table[h]].data == img.data):
duplicates1.append((idx, dup_table[h]))
#print 'Duplicate image: %d matches %d' % (idx, dup_table[h])
dup_table[h] = idx
overlaps = []
for idx,img in enumerate(images2):
h = hash(img.data)
if h in dup_table and (images1[dup_table[h]].data == img.data):
overlaps.append((dup_table[h], idx))
#print 'Test image %d is in the training set' % idx
return duplicates1, overlaps
def num_overlaps_with_diff_labels(overlap_indices, labels1, labels2):
count = 0
for olap in overlap_indices:
if labels1[olap[0]] != labels2[olap[1]]:
count += 1
return count
In [124]:
print( "\nMethod 1: hash and check equality (via sets)" )
t1 = time.time()
train_dups, train_valid_overlaps = find_dups_and_overlaps(train_dataset, valid_dataset)
test_dups, test_train_overlaps = find_dups_and_overlaps(test_dataset, train_dataset)
valid_dups, valid_test_overlaps = find_dups_and_overlaps(valid_dataset, test_dataset)
print( "train dups: %s, test_dups: %s, valid_dups: %s" % (len(train_dups), len(test_dups), len(valid_dups)) )
print( 'train/valid overlaps: %s, of which %s have different labels' % \
(len(train_valid_overlaps), num_overlaps_with_diff_labels(train_valid_overlaps, train_labels, valid_labels)) )
print( 'test/train overlaps: %s, of which %s have different labels' % \
(len(test_train_overlaps), num_overlaps_with_diff_labels(test_train_overlaps, test_labels, train_labels)) )
print( 'valid/test overlaps: %s, of which %s have different labels' % \
(len(valid_test_overlaps), num_overlaps_with_diff_labels(valid_test_overlaps, valid_labels, test_labels)) )
t2 = time.time()
print("Time: %0.2fs" % (t2 - t1))
In [ ]:
In [ ]:
In [129]:
counter = 0
whatisidx = []
whatisimg = []
for idx, img in enumerate(train_dataset):
if counter < 3:
whatisidx.append( idx)
whatisimg.append( img)
counter +=1
In [142]:
print(whatisidx); print(whatisimg[0].shape ) ;
whatisimg[0].max(), whatisimg[0].mean(), whatisimg[0].min(), whatisimg[0].std()
Out[142]:
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 [4]:
print( type(train_dataset)); print( train_dataset.dtype); print( type(train_labels)); print( train_labels.dtype)
print( train_dataset.shape); print( train_labels.shape);
print( train_labels.max(), train_labels.min(), train_labels.mean(), train_labels.std() )
In [32]:
import sklearn
from sklearn import linear_model
In [6]:
linlogreg = linear_model.LogisticRegression() # C = 1.0 seems to be the default
reshape training data set, i.e.
$ [0,1]^{N^x} \times [0,1]^{N^y} \subset \mathbb{R}^{N^x} \times \mathbb{R}^{N^y} \to [0,1]^{ N^x*N^y } \subset \mathbb{R}^{ N^x*N^y} $
In [19]:
train_dataset_lin = train_dataset.reshape( train_dataset.shape[0], train_dataset.shape[1]*train_dataset.shape[2])
print( len(train_dataset_lin))
In [26]:
test_dataset_lin = test_dataset.reshape( test_dataset.shape[0], test_dataset.shape[1]*test_dataset.shape[2])
print( len(test_dataset_lin))
valid_dataset_lin = valid_dataset.reshape( valid_dataset.shape[0], valid_dataset.shape[1]* valid_dataset.shape[2])
print( len( valid_dataset_lin))
In [18]:
import random
len(random.sample(range(len(train_labels)), 50))
Out[18]:
In [23]:
num_train_lst = [50,100,250,500,1000,2500,5000,10000]
indices_train_0 = random.sample(range(len(train_labels)), num_train_lst[0])
In [27]:
linlogreg.fit( train_dataset_lin[indices_train_0,:], train_labels[indices_train_0] )
Out[27]:
In [29]:
res_train = linlogreg.predict( train_dataset_lin)
print( type(res_train)); len(res_train)
Out[29]:
In [30]:
linlogreg.score( train_dataset_lin, train_labels)
Out[30]:
In [33]:
sklearn.metrics.accuracy_score( train_labels, res_train)
Out[33]:
In [34]:
linlogreg.get_params()
Out[34]:
In [35]:
print( linlogreg.score( test_dataset_lin, test_labels) );
print( linlogreg.score( valid_dataset_lin, valid_labels) );
In [ ]:
In [ ]:
In [ ]:
In [7]:
import os, sys
print( os.getcwd() ) # let's see where we are in the local directory
os.listdir( os.getcwd() )
Out[7]:
In [8]:
with open("notMNIST.pickle", 'rb') as f:
notMNISTpkl = pickle.load(f) # unpickle
In [10]:
print( type(notMNISTpkl) ); notMNISTpkl.keys()
Out[10]:
In [13]:
train_labels = notMNISTpkl["train_labels"]
train_dataset = notMNISTpkl["train_dataset"]
print( type(train_labels) ) ; print( type(train_dataset) )
In [25]:
test_labels = notMNISTpkl["test_labels"]
test_dataset = notMNISTpkl["test_dataset"]
valid_labels = notMNISTpkl["valid_labels"]
valid_dataset = notMNISTpkl["valid_dataset"]
In [ ]: