Merge pull request #10 from lucasb-eyer/mnist-revamp

Mnist revamp

Author: ikostrikov2

examples/MNIST/mnist.py

@@ -1,29 +1,29 @@
 import os
 import gzip
 import pickle
-import urllib
 import sys
+
+# Python 2/3 compatibility.
+try:
+    from urllib.request import urlretrieve
+except ImportError:
+    from urllib import urlretrieve
+
+
 '''Adapted from theano tutorial'''
 
 
-def load_mnist(data_file = './mnist.pkl.gz'):
+def load_mnist(data_file = os.path.join(os.path.dirname(__file__), 'mnist.pkl.gz')):
 
     if not os.path.exists(data_file):
-        origin = ('http://www.iro.umontreal.ca/~lisa/deep/data/mnist/mnist.pkl.gz')
-        print('Downloading data from %s' % origin)
-        urllib.urlretrieve(origin, data_file)
+        origin = 'http://www.iro.umontreal.ca/~lisa/deep/data/mnist/mnist.pkl.gz'
+        print('Downloading data from {}'.format(origin))
+        urlretrieve(origin, data_file)
 
     print('... loading data')
 
-    f = gzip.open(data_file, 'rb')
-    if sys.version_info[0] == 3:
-        train_set, valid_set, test_set = pickle.load(f, encoding='latin1')
-    else:
-        train_set, valid_set, test_set = pickle.load(f)
-    f.close()
-
-    train_set_x, train_set_y = train_set
-    valid_set_x, valid_set_y = valid_set
-    test_set_x, test_set_y = test_set
-
-    return (train_set_x, train_set_y), (valid_set_x, valid_set_y), (test_set_x, test_set_y)
+    with gzip.open(data_file, 'rb') as f:
+        if sys.version_info[0] == 3:
+            return pickle.load(f, encoding='latin1')
+        else:
+            return pickle.load(f)

examples/MNIST/run.py

@@ -18,8 +18,8 @@ def main(params):
 
     for epoch in range(100):
         model.training()
-        train(train_set_x, train_set_y, model, optimiser, criterion, epoch, params['batch_size'])
-        train(train_set_x, train_set_y, model, optimiser, criterion, epoch, params['batch_size'], 'stat')
+        train(train_set_x, train_set_y, model, optimiser, criterion, epoch, params['batch_size'], 'train')
+        train(train_set_x, train_set_y, model, optimiser, criterion, epoch, params['batch_size'], 'stats')
 
         model.evaluate()
         validate(test_set_x, test_set_y, model, epoch, params['batch_size'])
@@ -29,4 +29,4 @@ def main(params):
     params = {}
     params['lr'] = 0.1
     params['batch_size'] = 64
-    main(params)
+    main(params)

examples/MNIST/test.py

@@ -1,15 +1,14 @@
 import numpy as np
 from progress_bar import *
 import theano as _th
-from sklearn.metrics import accuracy_score
 
 def validate(dataset_x, dataset_y, model, epoch, batch_size):
     progress = make_progressbar('Testing', epoch, len(dataset_x))
     progress.start()
 
     mini_batch_input = np.empty(shape=(batch_size, 28*28), dtype=_th.config.floatX)
     mini_batch_targets = np.empty(shape=(batch_size, ), dtype=_th.config.floatX)
-    accuracy = 0
+    nerrors = 0
 
     for j in range((dataset_x.shape[0] + batch_size - 1) // batch_size):
         progress.update(j * batch_size)
@@ -24,7 +23,8 @@ def validate(dataset_x, dataset_y, model, epoch, batch_size):
             mini_batch_prediction.resize((dataset_x.shape[0] - j * batch_size, ))
             mini_batch_targets.resize((dataset_x.shape[0] - j * batch_size, ))
 
-        accuracy = accuracy + accuracy_score(mini_batch_targets, mini_batch_prediction, normalize=False)
+        nerrors += sum(mini_batch_targets != mini_batch_prediction)
 
     progress.finish()
-    print("Epoch #" + str(epoch) + ", Classification: " + str(float(accuracy) / dataset_x.shape[0] * 100.0))
+    accuracy = 1 - float(nerrors)/dataset_x.shape[0]
+    print("Epoch #{}, Classification accuracy: {:.2%} ({} errors)".format(epoch, accuracy, nerrors))

examples/MNIST/train.py

@@ -3,8 +3,8 @@
 import theano as _th
 
 
-def train(dataset_x, dataset_y, model, optimiser, criterion, epoch, batch_size, mode=None):
-    progress = make_progressbar('Training', epoch, len(dataset_x))
+def train(dataset_x, dataset_y, model, optimiser, criterion, epoch, batch_size, mode='train'):
+    progress = make_progressbar('Training ({})'.format(mode), epoch, len(dataset_x))
     progress.start()
 
     shuffle = np.random.permutation(len(dataset_x))
@@ -17,12 +17,14 @@ def train(dataset_x, dataset_y, model, optimiser, criterion, epoch, batch_size,
             mini_batch_input[k] = dataset_x[shuffle[j * batch_size + k]]
             mini_batch_targets[k] = dataset_y[shuffle[j * batch_size + k]]
 
-        if mode is None:
+        if mode == 'train':
             model.zero_grad_parameters()
             model.accumulate_gradients(mini_batch_input, mini_batch_targets, criterion)
             optimiser.update_parameters(model)
-        else:
+        elif mode == 'stats':
             model.accumulate_statistics(mini_batch_input)
+        else:
+            assert False, "Mode should be either 'train' or 'stats'"
 
         progress.update(j * batch_size)