Merge branch 'master' into patch-13

research/neural_programmer/wiki_data.py

@@ -22,6 +22,8 @@ columns.
 lookup answer (or matrix) is also split into number and word lookup matrix
 Author: aneelakantan (Arvind Neelakantan)
 """
+from __future__ import print_function
+
 import math
 import os
 import re
@@ -56,7 +58,7 @@ def correct_unicode(string):
   #string = re.sub("[“”«»]", "\"", string)
   #string = re.sub("[•†‡]", "", string)
   #string = re.sub("[‐‑–—]", "-", string)
-  string = re.sub(ur'[\u2E00-\uFFFF]', "", string)
+  string = re.sub(r'[\u2E00-\uFFFF]', "", string)
   string = re.sub("\\s+", " ", string).strip()
   return string
 
@@ -78,7 +80,7 @@ def full_normalize(string):
   # Remove trailing info in brackets
   string = re.sub("\[[^\]]*\]", "", string)
   # Remove most unicode characters in other languages
-  string = re.sub(ur'[\u007F-\uFFFF]', "", string.strip())
+  string = re.sub(r'[\u007F-\uFFFF]', "", string.strip())
   # Remove trailing info in parenthesis
   string = re.sub("\([^)]*\)$", "", string.strip())
   string = final_normalize(string)
@@ -207,7 +209,7 @@ class WikiQuestionGenerator(object):
     self.dev_loader = WikiQuestionLoader(dev_name, root_folder)
     self.test_loader = WikiQuestionLoader(test_name, root_folder)
     self.bad_examples = 0
-    self.root_folder = root_folder   
+    self.root_folder = root_folder
     self.data_folder = os.path.join(self.root_folder, "annotated/data")
     self.annotated_examples = {}
     self.annotated_tables = {}
@@ -298,7 +300,7 @@ class WikiQuestionGenerator(object):
             question_id, question, target_canon, context)
         self.annotated_tables[context] = []
       counter += 1
-    print "Annotated examples loaded ", len(self.annotated_examples)
+    print("Annotated examples loaded ", len(self.annotated_examples))
     f.close()
 
   def is_number_column(self, a):

research/next_frame_prediction/cross_conv/eval.py

@@ -20,6 +20,7 @@ import sys
 import time
 
 import numpy as np
+from six.moves import xrange
 import tensorflow as tf
 
 import model as cross_conv_model

research/next_frame_prediction/cross_conv/example_gen.py

@@ -18,6 +18,7 @@ import random
 import sys
 
 import numpy as np
+from six.moves import xrange
 import tensorflow as tf
 
 

research/next_frame_prediction/cross_conv/model.py

@@ -20,6 +20,7 @@ https://arxiv.org/pdf/1607.02586v1.pdf
 import math
 import sys
 
+from six.moves import xrange
 import tensorflow as tf
 
 slim = tf.contrib.slim

research/next_frame_prediction/cross_conv/reader.py

@@ -15,6 +15,7 @@
 
 """Read image sequence."""
 
+from six.moves import xrange
 import tensorflow as tf
 
 

research/next_frame_prediction/cross_conv/sprites_gen.py

@@ -21,6 +21,7 @@ import sys
 
 import numpy as np
 import scipy.misc
+from six.moves import xrange
 import tensorflow as tf
 
 

research/object_detection/dataset_tools/__init__.py

+

research/object_detection/dataset_tools/create_oid_tf_record.py

@@ -96,7 +96,7 @@ def main(_):
       tf_example = oid_tfrecord_creation.tf_example_from_annotations_data_frame(
           image_annotations, label_map, encoded_image)
       if tf_example:
-        shard_idx = long(image_id, 16) % FLAGS.num_shards
+        shard_idx = int(image_id, 16) % FLAGS.num_shards
         output_tfrecords[shard_idx].write(tf_example.SerializeToString())
 
 

research/object_detection/dataset_tools/create_pet_tf_record.py

@@ -160,8 +160,6 @@ def dict_to_tf_example(data,
     if not faces_only:
       mask_remapped = mask_np != 2
       masks.append(mask_remapped)
-    mask_stack = np.stack(masks).astype(np.float32)
-    masks_flattened = np.reshape(mask_stack, [-1])
 
   feature_dict = {
       'image/height': dataset_util.int64_feature(height),
@@ -184,8 +182,11 @@ def dict_to_tf_example(data,
       'image/object/view': dataset_util.bytes_list_feature(poses),
   }
   if not faces_only:
+    mask_stack = np.stack(masks).astype(np.float32)
+    masks_flattened = np.reshape(mask_stack, [-1])
     feature_dict['image/object/mask'] = (
         dataset_util.float_list_feature(masks_flattened.tolist()))
+
   example = tf.train.Example(features=tf.train.Features(feature=feature_dict))
   return example
 

research/object_detection/dataset_tools/oid_tfrecord_creation.py

@@ -18,6 +18,7 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
+from six.moves import xrange
 import tensorflow as tf
 
 from object_detection.core import standard_fields

research/object_detection/eval.py

@@ -103,7 +103,10 @@ def main(unused_argv):
 
   model_config = configs['model']
   eval_config = configs['eval_config']
-  input_config = configs['eval_input_config']
+  if FLAGS.eval_training_data:
+    input_config = configs['train_input_config']
+  else:
+    input_config = configs['eval_input_config']
 
   model_fn = functools.partial(
       model_builder.build,

research/object_detection/exporter.py

@@ -69,7 +69,7 @@ def freeze_graph_with_def_protos(
     if optimize_graph:
       logging.info('Graph Rewriter optimizations enabled')
       rewrite_options = rewriter_config_pb2.RewriterConfig(
-          optimize_tensor_layout=True)
+          layout_optimizer=rewriter_config_pb2.RewriterConfig.ON)
       rewrite_options.optimizers.append('pruning')
       rewrite_options.optimizers.append('constfold')
       rewrite_options.optimizers.append('layout')

research/object_detection/g3doc/exporting_models.md

@@ -8,7 +8,7 @@ graph proto. A checkpoint will typically consist of three files:
 * model.ckpt-${CHECKPOINT_NUMBER}.meta
 
 After you've identified a candidate checkpoint to export, run the following
-command from tensorflow/models/research/object_detection:
+command from tensorflow/models/research/:
 
 ``` bash
 # From tensorflow/models/research/

research/object_detection/g3doc/running_on_cloud.md

@@ -42,7 +42,7 @@ job using GPUs. A sample YAML file is given below:
 
 ```
 trainingInput:
-  runtimeVersion: "1.0"
+  runtimeVersion: "1.2"
   scaleTier: CUSTOM
   masterType: standard_gpu
   workerCount: 9
@@ -71,6 +71,7 @@ following command:
 ``` bash
 # From tensorflow/models/research/
 gcloud ml-engine jobs submit training object_detection_`date +%s` \
+    --runtime-version 1.2 \
     --job-dir=gs://${TRAIN_DIR} \
     --packages dist/object_detection-0.1.tar.gz,slim/dist/slim-0.1.tar.gz \
     --module-name object_detection.train \
@@ -90,6 +91,8 @@ Google Cloud Storage.
 Users can monitor the progress of their training job on the [ML Engine
 Dashboard](https://console.cloud.google.com/mlengine/jobs).
 
+Note: This sample is supported for use with 1.2 runtime version.
+
 ## Running an Evaluation Job on Cloud
 
 Evaluation jobs run on a single machine, so it is not necessary to write a YAML
@@ -98,6 +101,7 @@ job:
 
 ``` bash
 gcloud ml-engine jobs submit training object_detection_eval_`date +%s` \
+    --runtime-version 1.2 \
     --job-dir=gs://${TRAIN_DIR} \
     --packages dist/object_detection-0.1.tar.gz,slim/dist/slim-0.1.tar.gz \
     --module-name object_detection.eval \

research/object_detection/g3doc/running_pets.md

@@ -81,7 +81,7 @@ Oxford-IIIT Pet dataset into TFRecords. Run the following commands from the
 
 ``` bash
 # From tensorflow/models/research/
-python object_detection/create_pet_tf_record.py \
+python object_detection/dataset_tools/create_pet_tf_record.py \
     --label_map_path=object_detection/data/pet_label_map.pbtxt \
     --data_dir=`pwd` \
     --output_dir=`pwd`
@@ -203,12 +203,15 @@ For running the training Cloud ML job, we'll configure the cluster to use 10
 training jobs (1 master + 9 workers) and three parameters servers. The
 configuration file can be found at `object_detection/samples/cloud/cloud.yml`.
 
+Note: This sample is supported for use with 1.2 runtime version.
+
 To start training, execute the following command from the
 `tensorflow/models/research/` directory:
 
 ``` bash
 # From tensorflow/models/research/
 gcloud ml-engine jobs submit training `whoami`_object_detection_`date +%s` \
+    --runtime-version 1.2 \
     --job-dir=gs://${YOUR_GCS_BUCKET}/train \
     --packages dist/object_detection-0.1.tar.gz,slim/dist/slim-0.1.tar.gz \
     --module-name object_detection.train \
@@ -224,6 +227,7 @@ Once training has started, we can run an evaluation concurrently:
 ``` bash
 # From tensorflow/models/research/
 gcloud ml-engine jobs submit training `whoami`_object_detection_eval_`date +%s` \
+    --runtime-version 1.2 \
     --job-dir=gs://${YOUR_GCS_BUCKET}/train \
     --packages dist/object_detection-0.1.tar.gz,slim/dist/slim-0.1.tar.gz \
     --module-name object_detection.eval \

research/object_detection/object_detection_tutorial.ipynb

@@ -36,8 +36,8 @@
     "from matplotlib import pyplot as plt\n",
     "from PIL import Image\n",
     "\n",
-    "if tf.__version__ != '1.4.0':\n",
-    "  raise ImportError('Please upgrade your tensorflow installation to v1.4.0!')\n"
+    "if tf.__version__ < '1.4.0':\n",
+    "  raise ImportError('Please upgrade your tensorflow installation to v1.4.* or later!')\n"
    ]
   },
   {

research/object_detection/utils/np_box_list_ops.py

@@ -21,6 +21,7 @@ Example box operations that are supported:
 """
 
 import numpy as np
+from six.moves import xrange
 
 from object_detection.utils import np_box_list
 from object_detection.utils import np_box_ops

research/object_detection/utils/ops.py

@@ -203,9 +203,9 @@ def padded_one_hot_encoding(indices, depth, left_pad):
 
   TODO: add runtime checks for depth and indices.
   """
-  if depth < 0 or not isinstance(depth, (int, long) if six.PY2 else int):
+  if depth < 0 or not isinstance(depth, six.integer_types):
     raise ValueError('`depth` must be a non-negative integer.')
-  if left_pad < 0 or not isinstance(left_pad, (int, long) if six.PY2 else int):
+  if left_pad < 0 or not isinstance(left_pad, six.integer_types):
     raise ValueError('`left_pad` must be a non-negative integer.')
   if depth == 0:
     return None

research/object_detection/utils/visualization_utils_test.py

@@ -145,7 +145,7 @@ class VisualizationUtilsTest(tf.test.TestCase):
         for i in range(images_with_boxes_np.shape[0]):
           img_name = 'image_' + str(i) + '.png'
           output_file = os.path.join(self.get_temp_dir(), img_name)
-          print 'Writing output image %d to %s' % (i, output_file)
+          print('Writing output image %d to %s' % (i, output_file))
           image_pil = Image.fromarray(images_with_boxes_np[i, ...])
           image_pil.save(output_file)
 

research/pcl_rl/README.md

@@ -67,20 +67,27 @@ python trainer.py --logtostderr --batch_size=25 --env=HalfCheetah-v1 \
   --max_divergence=0.05 --value_opt=best_fit --critic_weight=0.0 \
 ```
 
-Run Mujoco task with Trust-PCL:
+To run Mujoco task using Trust-PCL (off-policy) use the below command.
+It should work well across all environments, given that you
+search sufficiently among
+
+(1) max_divergence (0.001, 0.0005, 0.002 are good values),
+
+(2) rollout (1, 5, 10 are good values),
+
+(3) tf_seed (need to average over enough random seeds).
 
 ```
 python trainer.py --logtostderr --batch_size=1 --env=HalfCheetah-v1 \
-  --validation_frequency=50 --rollout=10 --critic_weight=0.0 \
-  --gamma=0.995 --clip_norm=40 --learning_rate=0.002 \
-  --replay_buffer_freq=1 --replay_buffer_size=20000 \
-  --replay_buffer_alpha=0.1 --norecurrent --objective=pcl \
-  --max_step=100 --tau=0.0 --eviction=fifo --max_divergence=0.001 \
-  --internal_dim=64 --cutoff_agent=1000 \
-  --replay_batch_size=25 --nouse_online_batch --batch_by_steps \
-  --sample_from=target --value_opt=grad --value_hidden_layers=2 \
-  --update_eps_lambda --unify_episodes --clip_adv=1.0 \
-  --target_network_lag=0.99 --prioritize_by=step
+  --validation_frequency=250 --rollout=1 --critic_weight=1.0 --gamma=0.995 \
+  --clip_norm=40 --learning_rate=0.0001 --replay_buffer_freq=1 \
+  --replay_buffer_size=5000 --replay_buffer_alpha=0.001 --norecurrent \
+  --objective=pcl --max_step=10 --cutoff_agent=1000 --tau=0.0 --eviction=fifo \
+  --max_divergence=0.001 --internal_dim=256 --replay_batch_size=64 \
+  --nouse_online_batch --batch_by_steps --value_hidden_layers=2 \
+  --update_eps_lambda --nounify_episodes --target_network_lag=0.99 \
+  --sample_from=online --clip_adv=1 --prioritize_by=step --num_steps=1000000 \
+  --noinput_prev_actions --use_target_values --tf_seed=57
 ```
 
 Run Mujoco task with PCL constraint trust region: