cnn.patch

diff -u -r cnn_orig/convolutional_mlp.py cnn/convolutional_mlp.py
--- cnn_orig/convolutional_mlp.py	2014-08-08 12:15:47.412611962 +0900
+++ cnn/convolutional_mlp.py	2014-08-08 12:18:02.684614501 +0900
@@ -37,6 +37,18 @@
 from logistic_sgd import LogisticRegression, load_data
 from mlp import HiddenLayer
 
+from PIL import Image
+
+def pack(m,n,a):
+    r=numpy.zeros((n*28,m*28))
+    print a.shape,n*m
+    for k in xrange(min(n*m,a.shape[0])):
+        ii=28*(k//m)
+        jj=28*(k%m)
+        for i in xrange(28):
+            r[ii+i,jj:jj+28]=a[k,i*28:i*28+28]
+    return r.reshape(-1)
+
 
 class LeNetConvPoolLayer(object):
     """Pool Layer of a convolutional network """
@@ -129,6 +141,7 @@
     train_set_x, train_set_y = datasets[0]
     valid_set_x, valid_set_y = datasets[1]
     test_set_x, test_set_y = datasets[2]
+    valid_set = datasets[3]
 
     # compute number of minibatches for training, validation and testing
     n_train_batches = train_set_x.get_value(borrow=True).shape[0]
@@ -286,6 +299,21 @@
                 done_looping = True
                 break
 
+    validate_model2=theano.function(
+        [index],layer3.failures(y),
+        givens={
+            x: valid_set_x[index * batch_size: (index + 1) * batch_size],
+            y: valid_set_y[index * batch_size: (index + 1) * batch_size]})
+    nonzeros=[]
+    for i in xrange(n_valid_batches):
+        failures=validate_model2(i)
+        nz,=numpy.nonzero(failures)
+        nonzeros+=[j+i*batch_size for j in nz]
+    im=Image.new("1",(28*10,28*10))
+    sampled=nonzeros[:100]
+    im.putdata(1-pack(10,10,valid_set[0][sampled,:]),255.,0.)
+    im.show()
+
     end_time = time.clock()
     print('Optimization complete.')
     print('Best validation score of %f %% obtained at iteration %i,'\
diff -u -r cnn_orig/logistic_sgd.py cnn/logistic_sgd.py
--- cnn_orig/logistic_sgd.py	2014-08-08 12:15:47.412611962 +0900
+++ cnn/logistic_sgd.py	2014-08-08 12:18:52.308615433 +0900
@@ -143,6 +143,8 @@
         else:
             raise NotImplementedError()
 
+    def failures(self,y):
+        return T.neq(self.y_pred, y)
 
 def load_data(dataset):
     ''' Loads the dataset
@@ -212,7 +214,7 @@
     train_set_x, train_set_y = shared_dataset(train_set)
 
     rval = [(train_set_x, train_set_y), (valid_set_x, valid_set_y),
-            (test_set_x, test_set_y)]
+            (test_set_x, test_set_y), valid_set]
     return rval