Sync to latest.

Author: pkulzc

CODEOWNERS

@@ -17,6 +17,7 @@
 /research/inception/ @shlens @vincentvanhoucke
 /research/learned_optimizer/ @olganw @nirum
 /research/learning_to_remember_rare_events/ @lukaszkaiser @ofirnachum
+/research/learning_unsupervised_learning/ @lukemetz @nirum
 /research/lexnet_nc/ @vered1986 @waterson
 /research/lfads/ @jazcollins @susillo
 /research/lm_1b/ @oriolvinyals @panyx0718

official/utils/testing/scripts/presubmit.sh

@@ -53,12 +53,14 @@ py_test() {
 
 py2_test() {
   local PY_BINARY=$(which python2)
-  return $(py_test "${PY_BINARY}")
+  py_test "$PY_BINARY"
+  return $?
 }
 
 py3_test() {
   local PY_BINARY=$(which python3)
-  return $(py_test "${PY_BINARY}")
+  py_test "$PY_BINARY"
+  return $?
 }
 
 test_result=0

research/README.md

@@ -36,6 +36,8 @@ installation](https://www.tensorflow.org/install).
 -   [inception](inception): deep convolutional networks for computer vision.
 -   [learning_to_remember_rare_events](learning_to_remember_rare_events): a
     large-scale life-long memory module for use in deep learning.
+-   [learning_unsupervised_learning](learning_unsupervised_learning): a
+    meta-learned unsupervised learning update rule.
 -   [lexnet_nc](lexnet_nc): a distributed model for noun compound relationship
     classification.
 -   [lfads](lfads): sequential variational autoencoder for analyzing

research/compression/image_encoder/decoder.py

@@ -71,7 +71,7 @@ def main(_):
     return
 
   contents = ''
-  with tf.gfile.FastGFile(FLAGS.input_codes, 'r') as code_file:
+  with tf.gfile.FastGFile(FLAGS.input_codes, 'rb') as code_file:
     contents = code_file.read()
     loaded_codes = np.load(io.BytesIO(contents))
     assert ['codes', 'shape'] not in loaded_codes.files

research/compression/image_encoder/encoder.py

@@ -59,7 +59,7 @@ def main(_):
     print('\n--iteration must be between 0 and 15 inclusive.\n')
     return
 
-  with tf.gfile.FastGFile(FLAGS.input_image) as input_image:
+  with tf.gfile.FastGFile(FLAGS.input_image, 'rb') as input_image:
     input_image_str = input_image.read()
 
   with tf.Graph().as_default() as graph:

research/compression/image_encoder/msssim.py

@@ -199,9 +199,9 @@ def main(_):
     return
 
   with tf.gfile.FastGFile(FLAGS.original_image) as image_file:
-    img1_str = image_file.read()
+    img1_str = image_file.read('rb')
   with tf.gfile.FastGFile(FLAGS.compared_image) as image_file:
-    img2_str = image_file.read()
+    img2_str = image_file.read('rb')
 
   input_img = tf.placeholder(tf.string)
   decoded_image = tf.expand_dims(tf.image.decode_png(input_img, channels=3), 0)

research/learning_unsupervised_learning/.gitignore

@@ -0,0 +1 @@
+*.pyc

research/learning_unsupervised_learning/README.md

@@ -0,0 +1,37 @@
+# Learning Unsupervised Learning Rules
+This repository contains code and weights for the learned update rule
+presented in "Learning Unsupervised Learning Rules." At this time, this
+code can not meta-train the update rule.
+
+
+### Structure
+`run_eval.py` contains the main training loop. This constructs an op
+that runs one iteration of the learned update rule and assigns the
+results to variables. Additionally, it loads the weights from our
+pre-trained model.
+
+The base model and the update rule architecture definition can be found in
+`architectures/more_local_weight_update.py`. For a complete description
+of the model, see our [paper](https://arxiv.org/abs/1804.00222).
+
+### Dependencies
+[absl]([https://github.com/abseil/abseil-py), [tensorflow](https://tensorflow.org), [sonnet](https://github.com/deepmind/sonnet)
+
+### Usage
+
+First, download the [pre-trained optimizer model weights](https://storage.googleapis.com/learning_unsupervised_learning/200_tf_graph.zip) and extract it.
+
+```bash
+# move to the folder above this folder
+cd path_to/research/learning_unsupervised_learning/../
+
+# launch the eval script
+python -m learning_unsupervised_learning.run_eval \
+--train_log_dir="/tmp/learning_unsupervised_learning" \
+--checkpoint_dir="/path/to/downloaded/model/tf_graph_data.ckpt"
+```
+
+### Contact
+Luke Metz, Niru Maheswaranathan, Github: @lukemetz, @nirum. Email: {lmetz, nirum}@google.com
+
+

research/learning_unsupervised_learning/__init__.py

@@ -0,0 +1,17 @@
+# Copyright 2018 Google, Inc. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+
+import more_local_weight_update