Two bugfixes, improved err msg, and whitespace removal

gan_cifar.py

@@ -16,13 +16,20 @@
 import tflib.inception_score
 import tflib.plot
 
-# Download CIFAR-10 (Python version) at
-# https://www.cs.toronto.edu/~kriz/cifar.html and fill in the path to the
-# extracted files here!
-DATA_DIR = ''
+DATA_DIR = '/home/catherio/data/cifar10/cifar-10-batches-py/'
 if len(DATA_DIR) == 0:
-    raise Exception('Please specify path to data directory in gan_cifar.py!')
+    raise Exception('''
+Please specify path to data directory in gan_cifar.py!
 
+Download CIFAR-10 (Python version) at
+https://www.cs.toronto.edu/~kriz/cifar.html and fill in the path to the
+extracted files.
+
+> wget https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
+> tar -xvzf cifar-10-python.tar.gz
+''')
+
+DATASET = 'cifar' # (experimental) 'cifar' or 'svhn'
 MODE = 'wgan-gp' # Valid options are dcgan, wgan, or wgan-gp
 DIM = 128 # This overfits substantially; you're probably better off with 64
 LAMBDA = 10 # Gradient penalty lambda hyperparameter
@@ -110,7 +117,7 @@ def Discriminator(inputs):
         clip_bounds = [-.01, .01]
         clip_ops.append(
             tf.assign(
-                var, 
+                var,
                 tf.clip_by_value(var, clip_bounds[0], clip_bounds[1])
             )
         )
@@ -123,7 +130,7 @@ def Discriminator(inputs):
 
     # Gradient penalty
     alpha = tf.random_uniform(
-        shape=[BATCH_SIZE,1], 
+        shape=[BATCH_SIZE,1],
         minval=0.,
         maxval=1.
     )
@@ -167,17 +174,37 @@ def get_inception_score():
     all_samples = all_samples.reshape((-1, 3, 32, 32)).transpose(0,2,3,1)
     return lib.inception_score.get_inception_score(list(all_samples))
 
-# Dataset iterators
-train_gen, dev_gen = lib.cifar10.load(BATCH_SIZE, data_dir=DATA_DIR)
-def inf_train_gen():
-    while True:
-        for images,_ in train_gen():
-            yield images
 
 # Train loop
 with tf.Session() as session:
+    # Dataset iterators
+    if DATASET == 'cifar':
+        train_gen, dev_gen = lib.cifar10.load(BATCH_SIZE, data_dir=DATA_DIR)
+        def inf_gen(g):
+            while True:
+                for ims,_ in g:
+                    yield ims
+        train_gen = inf_gen(train_gen())
+        dev_gen = inf_gen(dev_gen())
+    elif DATASET == 'svhn':
+        from ganskill.svhn_data import input_fn
+        train_gen_tf = (input_fn(is_training=True,
+                                 batch_size=BATCH_SIZE)
+                        .make_one_shot_iterator())
+        dev_gen_tf = (input_fn(is_training=False,
+                               batch_size=BATCH_SIZE)
+                      .make_one_shot_iterator())
+
+        def dataset_iter(g_tf):
+            next_im, _ = g_tf.get_next()
+            reshape = tf.reshape(next_im, [-1, OUTPUT_DIM])
+            while True:
+                yield session.run(reshape)
+        train_gen = dataset_iter(train_gen_tf)
+        dev_gen = dataset_iter(dev_gen_tf)
+
     session.run(tf.initialize_all_variables())
-    gen = inf_train_gen()
+
 
     for iteration in xrange(ITERS):
         start_time = time.time()
@@ -190,7 +217,7 @@ def inf_train_gen():
         else:
             disc_iters = CRITIC_ITERS
         for i in xrange(disc_iters):
-            _data = gen.next()
+            _data = train_gen.next()
             _disc_cost, _ = session.run([disc_cost, disc_train_op], feed_dict={real_data_int: _data})
             if MODE == 'wgan':
                 _ = session.run(clip_disc_weights)
@@ -206,8 +233,8 @@ def inf_train_gen():
         # Calculate dev loss and generate samples every 100 iters
         if iteration % 100 == 99:
             dev_disc_costs = []
-            for images,_ in dev_gen():
-                _dev_disc_cost = session.run(disc_cost, feed_dict={real_data_int: images}) 
+            for images in dev_gen:
+                _dev_disc_cost = session.run(disc_cost, feed_dict={real_data_int: images})
                 dev_disc_costs.append(_dev_disc_cost)
             lib.plot.plot('dev disc cost', np.mean(dev_disc_costs))
             generate_image(iteration, _data)

tflib/inception_score.py

@@ -20,7 +20,7 @@
 DATA_URL = 'http://download.tensorflow.org/models/image/imagenet/inception-2015-12-05.tgz'
 softmax = None
 
-# Call this function with list of images. Each of elements should be a 
+# Call this function with list of images. Each of elements should be a
 # numpy array with values ranging from 0 to 255.
 def get_inception_score(images, splits=10):
   assert(type(images) == list)
@@ -32,7 +32,7 @@ def get_inception_score(images, splits=10):
   for img in images:
     img = img.astype(np.float32)
     inps.append(np.expand_dims(img, 0))
-  bs = 100
+  bs = 1
   with tf.Session() as sess:
     preds = []
     n_batches = int(math.ceil(float(len(inps)) / float(bs)))
@@ -88,9 +88,9 @@ def _progress(count, block_size, total_size):
                     new_shape.append(None)
                 else:
                     new_shape.append(s)
-            o._shape = tf.TensorShape(new_shape)
+            o.set_shape(tf.TensorShape(new_shape))
     w = sess.graph.get_operation_by_name("softmax/logits/MatMul").inputs[1]
-    logits = tf.matmul(tf.squeeze(pool3), w)
+    logits = tf.matmul(tf.squeeze(pool3, [1, 2]), w)
     softmax = tf.nn.softmax(logits)
 
 if softmax is None: