update summary API to latest

examples/DoReFa-Net/svhn-digit-dorefa.py

@@ -153,7 +153,7 @@ def get_config():
         global_step=get_global_step_var(),
         decay_steps=data_train.size() * 100,
         decay_rate=0.5, staircase=True, name='learning_rate')
-    tf.scalar_summary('learning_rate', lr)
+    tf.summary.scalar('lr', lr)
 
     return TrainConfig(
         dataset=data_train,

examples/GAN/DCGAN-CelebA.py

@@ -75,7 +75,7 @@ class Model(ModelDesc):
                 W_init=tf.truncated_normal_initializer(stddev=0.02)):
             with tf.variable_scope('gen'):
                 image_gen = self.generator(z)
-                tf.image_summary('gen', image_gen, max_images=30)
+                tf.summary.image('gen', image_gen, max_images=30)
             with tf.variable_scope('discrim'):
                 vecpos = self.discriminator(image_pos)
             with tf.variable_scope('discrim', reuse=True):

examples/GAN/GAN.py

@@ -52,8 +52,8 @@ class RandomZData(DataFlow):
 def build_GAN_losses(vecpos, vecneg):
     sigmpos = tf.sigmoid(vecpos)
     sigmneg = tf.sigmoid(vecneg)
-    tf.histogram_summary('sigmoid-pos', sigmpos)
-    tf.histogram_summary('sigmoid-neg', sigmneg)
+    tf.summary.histogram('sigmoid-pos', sigmpos)
+    tf.summary.histogram('sigmoid-neg', sigmneg)
 
     d_loss_pos = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(
         vecpos, tf.ones_like(vecpos)), name='d_CE_loss_pos')

examples/GAN/InfoGAN-mnist.py

@@ -66,7 +66,7 @@ class Model(ModelDesc):
                 W_init=tf.truncated_normal_initializer(stddev=0.02)):
             with tf.variable_scope('gen'):
                 image_gen = self.generator(z)
-                tf.image_summary('gen', image_gen, max_images=30)
+                tf.summary.image('gen', image_gen, max_images=30)
             with tf.variable_scope('discrim'):
                 vecpos, _ = self.discriminator(image_pos)
             with tf.variable_scope('discrim', reuse=True):

examples/SpatialTransformer/mnist-addition.py

@@ -61,7 +61,7 @@ class Model(ModelDesc):
         transform1 = tf.concat(1, [padded1[:,:,:,0], padded1[:,:,:,1]])
         transform2 = tf.concat(1, [padded2[:,:,:,0], padded2[:,:,:,1]])
         stacked = tf.concat(2, [img_orig, transform1, transform2], 'viz')
-        tf.image_summary('visualize',
+        tf.summary.image('visualize',
                 tf.expand_dims(stacked, -1), max_images=30)
 
         sampled = tf.concat(3, [sampled1, sampled2], 'sampled_concat')

examples/cifar-convnet.py

@@ -38,7 +38,7 @@ class Model(ModelDesc):
         keep_prob = tf.constant(0.5 if is_training else 1.0)
 
         if is_training:
-            tf.image_summary("train_image", image, 10)
+            tf.summary.image("train_image", image, 10)
 
         image = image / 4.0     # just to make range smaller
         with argscope(Conv2D, nl=BNReLU, use_bias=False, kernel_shape=3):

examples/mnist-convnet.py

@@ -104,7 +104,7 @@ def get_config():
 
     # This will also put the summary in tensorboard,stat.json and print in
     # terminal, but without the moving average
-    tf.scalar_summary('learning_rate', lr)
+    tf.summary.scalar('lr', lr)
 
     # get the config which contains everything necessary in a training
     return TrainConfig(

examples/svhn-digit-convnet.py

@@ -92,7 +92,7 @@ def get_config():
         global_step=get_global_step_var(),
         decay_steps=data_train.size() * 60,
         decay_rate=0.2, staircase=True, name='learning_rate')
-    tf.scalar_summary('learning_rate', lr)
+    tf.summary.scalar('lr', lr)
 
     return TrainConfig(
         dataset=data_train,

tensorpack/callbacks/stats.py

 # -*- coding: utf-8 -*-
-# File: stat.py
+# File: stats.py
 # Author: Yuxin Wu <ppwwyyxxc@gmail.com>
 
 import tensorflow as tf
@@ -40,6 +40,9 @@ class StatHolder(object):
         :param k: name
         :param v: value
         """
+        suffix = '-summary'
+        if k.endswith(suffix):
+            k = k[:-len(suffix)]
         self.stat_now[k] = float(v)
 
     def set_print_tag(self, print_tag):

tensorpack/tfutils/gradproc.py

@@ -110,7 +110,7 @@ class SummaryGradient(MapGradient):
         name = var.op.name
         if name not in _summaried_gradient:
             _summaried_gradient.add(name)
-            tf.histogram_summary(name + '/grad', grad)
+            tf.summary.histogram(name + '-grad', grad)
             add_moving_summary(rms(grad, name=name + '/rms'))
         return grad
 

tensorpack/tfutils/summary.py

@@ -38,11 +38,10 @@ def add_activation_summary(x, name=None):
         "Summary a scalar with histogram? Maybe use scalar instead. FIXME!"
     if name is None:
         name = x.name
-    with tf.name_scope('act_summary'):
-        tf.histogram_summary(name + '/activation', x)
-        tf.scalar_summary(name + '/activation_sparsity', tf.nn.zero_fraction(x))
-        tf.scalar_summary(
-                name + '/activation_rms', rms(x))
+    with tf.name_scope('activation-summary'):
+        tf.summary.histogram(name, x)
+        tf.summary.scalar(name + '-sparsity', tf.nn.zero_fraction(x))
+        tf.summary.scalar(name + '-rms', rms(x))
 
 def add_param_summary(summary_lists):
     """
@@ -59,25 +58,25 @@ def add_param_summary(summary_lists):
         name = var.name.replace(':0', '')
         if action == 'scalar':
             assert ndim == 0, "Scalar summary on high-dimension data. Maybe you want 'mean'?"
-            tf.scalar_summary(name, var)
+            tf.summary.scalar(name, var)
             return
         assert ndim > 0, "Cannot perform {} summary on scalar data".format(action)
         if action == 'histogram':
-            tf.histogram_summary(name, var)
+            tf.summary.histogram(name, var)
             return
         if action == 'sparsity':
-            tf.scalar_summary(name + '/sparsity', tf.nn.zero_fraction(var))
+            tf.summary.scalar(name + '-sparsity', tf.nn.zero_fraction(var))
             return
         if action == 'mean':
-            tf.scalar_summary(name + '/mean', tf.reduce_mean(var))
+            tf.summary.scalar(name + '-mean', tf.reduce_mean(var))
             return
         if action == 'rms':
-            tf.scalar_summary(name + '/rms', rms(var))
+            tf.summary.scalar(name + '-rms', rms(var))
             return
         raise RuntimeError("Unknown summary type: {}".format(action))
 
     params = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES)
-    with tf.name_scope('param_summary'):
+    with tf.name_scope('param-summary'):
         for p in params:
             name = p.name
             for rgx, actions in summary_lists:
@@ -111,7 +110,7 @@ def summary_moving_average(tensors=None):
     """
     if tensors is None:
         tensors = tf.get_collection(MOVING_SUMMARY_VARS_KEY)
-    with tf.name_scope('EMA_summary'):
+    with tf.name_scope('EMA-summary'):
         # TODO will produce EMA_summary/tower0/xxx. not elegant
         with tf.name_scope(None):
             averager = tf.train.ExponentialMovingAverage(
@@ -119,6 +118,6 @@ def summary_moving_average(tensors=None):
         avg_maintain_op = averager.apply(tensors)
         for idx, c in enumerate(tensors):
             name = re.sub('tower[p0-9]+/', '', c.op.name)
-            tf.scalar_summary(name, averager.average(c))
+            tf.summary.scalar(name, averager.average(c))
         return avg_maintain_op
 

tensorpack/tfutils/symbolic_functions.py

@@ -116,5 +116,6 @@ def get_scalar_var(name, init_value, summary=False, trainable=False):
             initializer=tf.constant_initializer(init_value),
             trainable=trainable)
     if summary:
-        tf.scalar_summary(name, ret)
+        # this is recognized in callbacks.StatHolder
+        tf.summary.scalar(name + '-summary', ret)
     return ret

tensorpack/train/base.py

@@ -109,8 +109,8 @@ class Trainer(object):
 
         if not hasattr(logger, 'LOG_DIR'):
             raise RuntimeError("logger directory wasn't set!")
-        self.summary_writer = tf.train.SummaryWriter(logger.LOG_DIR, graph=self.sess.graph)
-        self.summary_op = tf.merge_all_summaries()
+        self.summary_writer = tf.summary.FileWriter(logger.LOG_DIR, graph=self.sess.graph)
+        self.summary_op = tf.summary.merge_all()
         # create an empty StatHolder
         self.stat_holder = StatHolder(logger.LOG_DIR)