misc fix

examples/DoReFa-Net/alexnet-dorefa.py

@@ -38,7 +38,7 @@ Accuracy:
     With (W,A,G)=(1,2,4), 63% error.
 
 Speed:
-    About 3.5 iteration/s on 4 Tesla M40. (Each epoch is set to 10000 iterations)
+    About 2.8 iteration/s on 1 TitanX. (Each epoch is set to 10000 iterations)
 
 To Train:
     ./alexnet-dorefa.py --dorefa 1,2,6 --data PATH --gpu 0,1,2,3
@@ -66,8 +66,7 @@ BITW = 1
 BITA = 2
 BITG = 6
 TOTAL_BATCH_SIZE = 128
-NUM_GPU = 4
-BATCH_SIZE = TOTAL_BATCH_SIZE // NUM_GPU
+BATCH_SIZE = 64
 
 class Model(ModelDesc):
     def _get_input_vars(self):
@@ -301,6 +300,10 @@ if __name__ == '__main__':
         run_image(Model(), ParamRestore(np.load(args.load, encoding='latin1').item()), args.run)
         sys.exit()
 
+    assert args.gpu is not None, "Need to specify a list of gpu for training!"
+    NR_GPU = len(args.gpu.split(','))
+    BATCH_SIZE = TOTAL_BATCH_SIZE // NR_GPU
+
     config = get_config()
     if args.load:
         config.session_init = SaverRestore(args.load)

examples/GAN/GAN.py

@@ -54,16 +54,16 @@ def build_GAN_losses(vecpos, vecneg):
     tf.histogram_summary('sigmoid-neg', sigmneg)
 
     d_loss_pos = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(
-        vecpos, tf.ones_like(vecpos)), name='d_loss_pos')
+        vecpos, tf.ones_like(vecpos)), name='d_CE_loss_pos')
     d_loss_neg = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(
-        vecneg, tf.zeros_like(vecneg)), name='d_loss_neg')
+        vecneg, tf.zeros_like(vecneg)), name='d_CE_loss_neg')
 
     d_pos_acc = tf.reduce_mean(tf.cast(sigmpos > 0.5, tf.float32), name='pos_acc')
     d_neg_acc = tf.reduce_mean(tf.cast(sigmneg < 0.5, tf.float32), name='neg_acc')
 
     g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(
-        vecneg, tf.ones_like(vecneg)), name='g_loss')
-    d_loss = tf.add(d_loss_pos, d_loss_neg, name='d_loss')
+        vecneg, tf.ones_like(vecneg)), name='g_CE_loss')
+    d_loss = tf.add(d_loss_pos, d_loss_neg, name='d_CE_loss')
     add_moving_summary(d_loss_pos, d_loss_neg,
             g_loss, d_loss,
             d_pos_acc, d_neg_acc)

examples/ResNet/imagenet-resnet.py

 #!/usr/bin/env python
 # -*- coding: UTF-8 -*-
-# File: imagenet-resnet-short.py
+# File: imagenet-resnet.py
 # Author: Yuxin Wu <ppwwyyxx@gmail.com>
 
 import cv2
@@ -233,7 +233,7 @@ if __name__ == '__main__':
 
     assert args.gpu is not None, "Need to specify a list of gpu for training!"
     NR_GPU = len(args.gpu.split(','))
-    BATCH_SIZE = TOTAL_BATCH_SIZE / NR_GPU
+    BATCH_SIZE = TOTAL_BATCH_SIZE // NR_GPU
 
     logger.auto_set_dir()
     config = get_config()

tensorpack/models/regularize.py

@@ -40,8 +40,12 @@ def regularize_cost(regex, func, name=None):
 
 
 @layer_register(log_shape=False)
-def Dropout(x, keep_prob=0.5):
-    is_training = get_current_tower_context().is_training
+def Dropout(x, keep_prob=0.5, is_training=None):
+    """
+    :param is_training: if None, will use the current context by default.
+    """
+    if is_training is None:
+        is_training = get_current_tower_context().is_training
     keep_prob = tf.constant(keep_prob if is_training else 1.0)
     return tf.nn.dropout(x, keep_prob)
 

tensorpack/utils/fs.py

@@ -5,6 +5,7 @@
 
 import os, sys
 from six.moves import urllib
+import errno
 from . import logger
 
 __all__ = ['mkdir_p', 'download']
@@ -17,7 +18,7 @@ def mkdir_p(dirname):
     try:
         os.makedirs(dirname)
     except OSError as e:
-        if e.errno != 17:
+        if e.errno != errno.EEXIST:
             raise e