TF-Seq2Seq via Docker

README.md

@@ -1,3 +1,5 @@
+# TF-Seq2Seq (v0.1.0)
+
 [![CircleCI](https://circleci.com/gh/google/seq2seq.svg?style=svg)](https://circleci.com/gh/google/seq2seq)
 
 ---

docs/getting_started.md

@@ -1,7 +1,7 @@
 ## Download & Setup
 
 To use tf-seq2seq you need a working installation of TensorFlow 1.0 with
-Python 2.7 or Python 3.5. Follow the [TensorFlow Getting Started](https://www.tensorflow.org/versions/r1.0/get_started/os_setup) guide for detailed setup instructions. With TensorFlow installed, you can clone this repository:
+Python 2.7 or Python 3.5. Follow the [Installing TensorFlow](https://www.tensorflow.org/install/) guide for detailed setup instructions. With TensorFlow installed, you can clone this repository:
 
 ```bash
 git clone https://github.com/google/seq2seq.git
@@ -11,12 +11,96 @@ cd seq2seq
 pip install -e .
 ```
 
+## TF-Seq2Seq via Docker
+
+Now is possible to run [tf-seq2seq via Docker](https://github.com/google/seq2seq/seq2seq/tools/docker). Unfortunately, we haven't deployed an automatic pipeline to build container, so we provide the Dockerfile to build images for your repository.
+
+### Build Image
+First of all you have to build the image for running tf-seq2seq, according to your python preference(python2.7 or python3.5) and hardware capabilities(with or without GPU) from the right Dockerfile
+
+* [Docker for Python 2.7](https://github.com/google/seq2seq/seq2seq/tools/docker/py27)
+* [Docker for Python 3.5](https://github.com/google/seq2seq/seq2seq/tools/docker/py35)
+
+Both the directories have Dockerfile to build image with basic depenecies(Dockerfile), development dependecies(Dockerfile.devel) and GPU support(Dockerfile.gpu and Dockerfile.devel-gpu). The file provided, are coded from the [official TensorFlow Docker directory](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/tools/docker/README.md) with **latest** TensorFlow version/tags.
+Now you are ready to build the image(you have to be in py27 or py35 dir to run this command):
+
+```bash
+$ docker build -t image_name -f Dockerfile.suffix .
+```
+
+We give you some examples:
+```bash
+# In py27 dir
+$ docker build -t py27 -f Dockerfile .
+$ docker build -t py27-devel-gpu -f Dockerfile.devel-gpu .
+
+# In py35 dir
+$ docker build -t py35-devel -f Dockerfile.devel .
+$ docker build -t py35-gpu -f Dockerfile.gpu .
+```
+
+### Build Custom Image
+To build a container with a certain TensorFlow version, look at the tags/version in the [Docker Store](https://store.docker.com/community/images/tensorflow/tensorflow/tags), then change the first line's tags of the Dockerfile with your choice and run the `docker build` command.
+
+Some examples:
+```
+# In py27 Dockerfile, with the TensorFlow 1.0.1 version
+FROM tensorflow/tensorflow:latest --Became--> FROM tensorflow/tensorflow:1.0.1
+
+#In py35 Dockerfile.devel-gpu, with the TensorFlow 1.1.0-rc2 version
+FROM tensorflow/tensorflow:latest-devel-gpu-py3 --Became-->	1.2.0-rc2-devel-gpu-py3
+```
+
+Once a container is built, you will find the Tf-Seq2Seq package in the `/src/seq2seq` path(the default workdir).
+
+### Running container
+
+Run non-GPU container using
+
+    $ docker run -it -p hostPort:containerPort image_name(provided during the building step)
+
+Some examples
+```bash
+# Run a container with the Tf-Seq2Seq package in a py27 developer env
+$ docker run -it py27-devel
+
+# Run a container with the Tf-Seq2Seq package in a py35 env and look at the result with TensorBoard
+$ docker run -it -p 6006:6006 py35
+
+# Run a container with the Tf-Seq2Seq package in a py27 env and work at a Tf-seq2seq package cloned in the Host through the container
+$ docker run -it -v $(pwd):/seq2seq -w /seq2seq py27
+```
+
+For GPU support install NVidia drivers (ideally latest) and
+[nvidia-docker](https://github.com/NVIDIA/nvidia-docker). Run using
+
+    $ nvidia-docker run -it -p hostPort:containerPort {repository_name}(provided during the building step)
+
+The examples are the same as above with the only difference that the command is `nvidia-docker`.
+
+Note: If you would have a problem running nvidia-docker you may try the old method
+we have used. But it is not recommended. If you find a bug in nvidia-docker, please report
+it there and try using nvidia-docker as described above.
+
+```bash
+    $ export CUDA_SO=$(\ls /usr/lib/x86_64-linux-gnu/libcuda.* | xargs -I{} echo '-v {}:{}')
+    $ export DEVICES=$(\ls /dev/nvidia* | xargs -I{} echo '--device {}:{}')
+    $ docker run -it -p 8888:8888 $CUDA_SO $DEVICES gcr.io/tensorflow/tensorflow:latest-gpu
+```
+
+## Validate your installation
+
 To make sure everything works as expect you can run a simple pipeline unit test:
 
 ```bash
 python -m unittest seq2seq.test.pipeline_test
 ```
 
+Or the full test pipeline(you have to download nose if you haven't installed the tf-seq2seq via Docker):
+```bash
+nosetests
+```
+
 If you see a "OK" message, you are all set. Note that you may need to install pyrouge, pyyaml, and matplotlib, in order for these tests to pass. If you run into other setup issues,
 please [file a Github issue](https://github.com/google/seq2seq/issues).
 

docs/nmt.md

@@ -255,7 +255,7 @@ The above command also demonstrates how to pass several tasks to the inference s
 
 ### Evaluating specific checkpoint
 
-The training script will save multiple model checkpoints throughout training. The exact checkpoint behavior can be controlled via [training script flags](training/). By default, the inference script evaluates the latest checkpoint in the model directory. To evaluate a specific checkpiint you can pass the `checkpoint_path` flag.
+The training script will save multiple model checkpoints throughout training. The exact checkpoint behavior can be controlled via [training script flags](training/). By default, the inference script evaluates the latest checkpoint in the model directory. To evaluate a specific checkpoint you can pass the `checkpoint_path` flag.
 
 
 ## Calcuating BLEU scores

pylintrc

@@ -292,7 +292,7 @@ generated-members=set_shape,np.float32
 # List of decorators that produce context managers, such as
 # contextlib.contextmanager. Add to this list to register other decorators that
 # produce valid context managers.
-contextmanager-decorators=contextlib.contextmanager
+contextmanager-decorators=contextlib.contextmanager,tensorflow.python.util.tf_contextlib.contextmanager
 
 
 [VARIABLES]

seq2seq/contrib/seq2seq/helper.py

@@ -32,8 +32,13 @@
 
 import six
 
-from tensorflow.contrib.distributions.python.ops import bernoulli
-from tensorflow.contrib.distributions.python.ops import categorical
+try:
+  from tensorflow.python.ops.distributions import bernoulli
+  from tensorflow.python.ops.distributions import categorical
+except:
+  # Backwards compatibility with TensorFlow prior to 1.2.
+  from tensorflow.contrib.distributions.python.ops import bernoulli
+  from tensorflow.contrib.distributions.python.ops import categorical
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
 from tensorflow.python.layers import base as layers_base

seq2seq/data/input_pipeline.py

@@ -29,7 +29,9 @@
 import six
 
 import tensorflow as tf
+# pylint: disable=no-name-in-module
 from tensorflow.contrib.slim.python.slim.data import tfexample_decoder
+# pylint: enable=no-name-in-module
 
 from seq2seq.configurable import Configurable
 from seq2seq.data import split_tokens_decoder, parallel_data_provider

seq2seq/data/parallel_data_provider.py

@@ -22,8 +22,10 @@
 import numpy as np
 
 import tensorflow as tf
+# pylint: disable=no-name-in-module
 from tensorflow.contrib.slim.python.slim.data import data_provider
 from tensorflow.contrib.slim.python.slim.data import parallel_reader
+# pylint: enable=no-name-in-module
 
 from seq2seq.data import split_tokens_decoder
 

seq2seq/data/sequence_example_decoder.py

@@ -14,7 +14,9 @@
 """A decoder for tf.SequenceExample"""
 
 import tensorflow as tf
+# pylint: disable=no-name-in-module
 from tensorflow.contrib.slim.python.slim.data import data_decoder
+# pylint: enable=no-name-in-module
 
 
 class TFSEquenceExampleDecoder(data_decoder.DataDecoder):

seq2seq/data/split_tokens_decoder.py

@@ -21,7 +21,9 @@
 from __future__ import unicode_literals
 
 import tensorflow as tf
+# pylint: disable=no-name-in-module
 from tensorflow.contrib.slim.python.slim.data import data_decoder
+# pylint: enable=no-name-in-module
 
 
 class SplitTokensDecoder(data_decoder.DataDecoder):

seq2seq/encoders/image_encoder.py

@@ -20,8 +20,10 @@
 from __future__ import print_function
 
 import tensorflow as tf
+# pylint: disable=no-name-in-module
 from tensorflow.contrib.slim.python.slim.nets.inception_v3 \
   import inception_v3_base
+# pylint: enable=no-name-in-module
 
 from seq2seq.encoders.encoder import Encoder, EncoderOutput
 

seq2seq/encoders/rnn_encoder.py

@@ -21,7 +21,6 @@
 
 import copy
 import tensorflow as tf
-from tensorflow.contrib.rnn.python.ops import rnn
 
 from seq2seq.encoders.encoder import Encoder, EncoderOutput
 from seq2seq.training import utils as training_utils
@@ -186,7 +185,7 @@ def encode(self, inputs, sequence_length, **kwargs):
     cells_fw = _unpack_cell(cell_fw)
     cells_bw = _unpack_cell(cell_bw)
 
-    result = rnn.stack_bidirectional_dynamic_rnn(
+    result = tf.contrib.rnn.stack_bidirectional_dynamic_rnn(
         cells_fw=cells_fw,
         cells_bw=cells_bw,
         inputs=inputs,

seq2seq/metrics/metric_specs.py

@@ -28,7 +28,9 @@
 
 import tensorflow as tf
 from tensorflow.contrib import metrics
+# pylint: disable=no-name-in-module
 from tensorflow.contrib.learn import MetricSpec
+# pylint: enable=no-name-in-module
 
 from seq2seq.data import postproc
 from seq2seq.configurable import Configurable

seq2seq/test/hooks_test.py

@@ -39,16 +39,22 @@ class TestPrintModelAnalysisHook(tf.test.TestCase):
   def test_begin(self):
     model_dir = tempfile.mkdtemp()
     outfile = tempfile.NamedTemporaryFile()
-    tf.get_variable("weigths", [128, 128])
+    tf.get_variable("weights", [128, 128])
     hook = hooks.PrintModelAnalysisHook(
         params={}, model_dir=model_dir, run_config=tf.contrib.learn.RunConfig())
     hook.begin()
 
     with gfile.GFile(os.path.join(model_dir, "model_analysis.txt")) as file:
       file_contents = file.read().strip()
 
-    self.assertEqual(file_contents.decode(), "_TFProfRoot (--/16.38k params)\n"
-                     "  weigths (128x128, 16.38k/16.38k params)")
+    lines = tf.compat.as_text(file_contents).split("\n")
+    if len(lines) == 3:
+      # TensorFlow v1.2 includes an extra header line
+      self.assertEqual(lines[0], "node name | # parameters")
+
+    self.assertEqual(lines[-2], "_TFProfRoot (--/16.38k params)")
+    self.assertEqual(lines[-1], "  weights (128x128, 16.38k/16.38k params)")
+
     outfile.close()
 
 
@@ -125,7 +131,7 @@ def tearDown(self):
   def test_capture(self):
     global_step = tf.contrib.framework.get_or_create_global_step()
     # Some test computation
-    some_weights = tf.get_variable("weigths", [2, 128])
+    some_weights = tf.get_variable("weights", [2, 128])
     computation = tf.nn.softmax(some_weights)
 
     hook = hooks.MetadataCaptureHook(

seq2seq/tools/docker/README.md

@@ -0,0 +1,100 @@
+# Using TF-Seq2Seq via Docker
+
+This directory contains `Dockerfile`s to make it easy to get up and running with
+TF-Seq2Seq via [Docker](https://www.docker.com/).
+
+
+## Installing Docker
+
+General installation instructions are
+[on the Docker site](https://docs.docker.com/), but we give some
+quick links here:
+
+* [Docker for Mac OSX](https://www.docker.com/docker-mac)
+* [Docker for Windows PC](https://www.docker.com/docker-windows)
+* [Docker for Debian](https://www.docker.com/docker-debian)
+* [Docker for Ubuntu](https://www.docker.com/docker-ubuntu)
+* [Docker for CentOS Distribution](https://www.docker.com/docker-centos-distribution)
+
+
+## Dockerfile
+
+At the moment we haven't deployed an automatic pipeline to build container, so we provide the Dockerfile to build images
+This directory is structured to maintain Dockerfile for python2.7(py27) and python3.5(py35).
+
+* `Dockerfile` - TF-Seq2Seq - CPU only!
+
+* `Dockerfile.devel` - Developer build for TF-Seq2Seq - CPU only!
+
+* `Dockerfile.gpu` - TF-Seq2Seq with support of NVidia CUDA
+
+* `Dockerfile.devel-gpu` - Developer build for TF-Seq2Seq with support of NVidia CUDA
+
+The file provided, are coded from the [official TensorFlow Docker directory](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/tools/docker/README.md) and with **latest** TensorFlow version/tags.
+
+
+## Build image
+
+To build the container's image run in the directory of your favorite python version(some_path/seq2seq/seq2seq/tools/docker/{python-version}/):
+
+	$ docker build -t image_name -f Dockerfile.suffix .
+
+
+Some examples
+```bash
+# In py27 dir
+$ docker build -t py27 -f Dockerfile .
+$ docker build -t py27-devel-gpu -f Dockerfile.devel-gpu .
+
+# In py35 dir
+$ docker build -t py35-devel -f Dockerfile.devel .
+$ docker build -t py35-gpu -f Dockerfile.gpu .
+```
+
+To build a container with a certain TensorFlow version, look at the tags/version in the [Docker Store](https://store.docker.com/community/images/tensorflow/tensorflow/tags), then change the first line's tags of the Dockerfile with your choice:
+
+Some examples:
+```
+# In py27 Dockerfile, with the TensorFlow 1.0.1 version
+FROM tensorflow/tensorflow:latest --Became--> FROM tensorflow/tensorflow:1.0.1
+
+#In py35 Dockerfile.devel-gpu, with the TensorFlow 1.1.0-rc2 version
+FROM tensorflow/tensorflow:latest-devel-gpu-py3 --Became-->	1.2.0-rc2-devel-gpu-py3
+```
+
+Once a container is built, you will find the Tf-Seq2Seq package in the `/src/seq2seq` path(the default workdir).
+
+## Running container
+
+Run non-GPU container using
+
+    $ docker run -it -p hostPort:containerPort image_name(provided during the building step)
+
+Some examples
+```bash
+# Run a container with the Tf-Seq2Seq package in a py27 developer env
+$ docker run -it py27-devel
+
+# Run a container with the Tf-Seq2Seq package in a py35 env and look at the result with TensorBoard
+$ docker run -it -p 6006:6006 py35
+
+# Run a container with the Tf-Seq2Seq package in a py27 env and work at a Tf-seq2seq package cloned in the Host through the container
+$ docker run -it -v $(pwd):/seq2seq -w /seq2seq py27
+```
+
+For GPU support install NVidia drivers (ideally latest) and
+[nvidia-docker](https://github.com/NVIDIA/nvidia-docker). Run using
+
+    $ nvidia-docker run -it -p hostPort:containerPort {repository_name}(provided during the building step)
+
+The examples are the same as above with the only difference that the command is `nvidia-docker`.
+
+Note: If you would have a problem running nvidia-docker you may try the old method
+we have used. But it is not recommended. If you find a bug in nvidia-docker, please report
+it there and try using nvidia-docker as described above.
+
+```bash
+    $ export CUDA_SO=$(\ls /usr/lib/x86_64-linux-gnu/libcuda.* | xargs -I{} echo '-v {}:{}')
+    $ export DEVICES=$(\ls /dev/nvidia* | xargs -I{} echo '--device {}:{}')
+    $ docker run -it -p 8888:8888 $CUDA_SO $DEVICES gcr.io/tensorflow/tensorflow:latest-gpu
+```

seq2seq/tools/docker/py27/Dockerfile

@@ -0,0 +1,31 @@
+FROM tensorflow/tensorflow:latest
+
+MAINTAINER Alessio Gozzoli <[email protected]>
+
+RUN apt-get update && apt-get install -y \
+			python-tk \
+			python3-tk \
+			git \
+			&& \
+		apt-get clean && \
+		rm -rf /var/lib/apt/lists/*
+
+# Install Seq2Seq Dependencies
+WORKDIR /
+RUN pip --no-cache-dir install -e git+https://github.com/google/seq2seq.git#egg=seq2seq && \
+    pip --no-cache-dir install \
+    			nose \
+    			pylint \
+    			tox \
+    			yapf \
+    			mkdocs
+
+# Set Matploblib Backend
+RUN mkdir -p /root/.config/matplotlib/ && \
+			touch /root/.config/matplotlib/matplotlibrc && \
+			echo "backend : Agg" >> /root/.config/matplotlib/matplotlibrc
+
+# Default workdir
+WORKDIR "/src/seq2seq"
+
+CMD ["/bin/bash"]