Skip to content

Latest commit

 

History

History
342 lines (296 loc) · 11.3 KB

README_api.md

File metadata and controls

342 lines (296 loc) · 11.3 KB
Raw

  • how to run trained syntaxnet model using tensorflow serving?

  • i copied parsey-mcparseface-api to api directory and modify little bit. and follow xtknight instructions.

  • server

# bazel versions : 0.2.2b ( see : https://github.com/dsindex/syntaxnet/issues/17 )

$ git clone https://github.com/dsindex/syntaxnet.git work
$ cd work
$ git clone --recurse-submodules https://github.com/tensorflow/serving
# checkout proper version of serving
$ cd serving
$ git checkout 89e9dfbea055027bc31878ee8da66b54a701a746
$ git submodule update --init --recursive
# checkout proper version of tf_models
$ cd tf_models
$ git checkout a4b7bb9a5dd2c021edcd3d68d326255c734d0ef0

# you can use the tf_models/syntaxnet to build your own model and serve it.
# note that if you have an error(-fno-canonical-system-headers) while compiling syntaxnet on OS X Sierra, 
# modify tf_models/syntaxnet/tensorflow/tensorflow/workspace.bzl file :
# native.git_repository(
#    name = "re2",
#    remote = "https://github.com/google/re2.git",
#    commit = "ae9cb49",
#  )

# apply patch by dmansfield to serving/tf_models/syntaxnet 
$ patch -p1 < ../../api/pr250-patch-a4b7bb9a.diff.txt
$ cd ../

# configure serving/tensorflow
$ cd tensorflow
$ ./configure
$ cd ../../

# modify serving/tensorflow_serving/workspace.bzl for referencing syntaxnet
$ cp api/modified_workspace.bzl serving/tensorflow_serving/workspace.bzl
$ cat api/modified_workspace.bzl
#  ... 
#  native.local_repository(
#    name = "syntaxnet",
#    path = workspace_dir + "/tf_models/syntaxnet",
#  )
#  ...

# append build instructions to serving/tensorflow_serving/example/BUILD
$ cat api/append_BUILD >> serving/tensorflow_serving/example/BUILD

# copy parsey_api.cc, parsey_api.proto to example directory to build
$ cp api/parsey_api* serving/tensorflow_serving/example/

# build parsey_api 
$ cd serving
# please check bazel version == 0.2.2b
$ bazel --output_user_root=bazel_root build --nocheck_visibility -c opt -s //tensorflow_serving/example:parsey_api --genrule_strategy=standalone --spawn_strategy=standalone --verbose_failures

# if you have a trouble on downloading zlib ( https://github.com/tensorflow/tensorflow/issues/6668 )
# modify bellow files :
# ./tf_models/syntaxnet/WORKSPACE
# ./tensorflow/tensorflow/workspace.bzl
# ./tf_models/syntaxnet/tensorflow/tensorflow/workspace.bzl
# see : https://github.com/tensorflow/tensorflow/issues/6594
# 1. ./tf_models/syntaxnet/WORKSPACE
# new_http_archive(
#      name = "ts_zlib_archive",
#      url = "http://zlib.net/fossils/zlib-1.2.8.tar.gz",
#      sha256 = "36658cb768a54c1d4dec43c3116c27ed893e88b02ecfcb44f2166f9c0b7f2a0d",
#      strip_prefix = "zlib-1.2.8",
#      build_file = "zlib.BUILD",
# )
# 2. ./tensorflow/tensorflow/workspace.bzl
# native.new_http_archive(
#    name = "zlib_archive",
#    url = "http://zlib.net/fossils/zlib-1.2.8.tar.gz",
#    sha256 = "36658cb768a54c1d4dec43c3116c27ed893e88b02ecfcb44f2166f9c0b7f2a0d",
#    build_file = path_prefix + "zlib.BUILD",
#  )
# 3. ./tf_models/syntaxnet/tensorflow/tensorflow/workspace.bzl
# native.new_http_archive(
#      name = "zlib_archive",
#      url = "http://zlib.net/fossils/zlib-1.2.8.tar.gz",
#      sha256 = "36658cb768a54c1d4dec43c3116c27ed893e88b02ecfcb44f2166f9c0b7f2a0d",
#      strip_prefix = "zlib-1.2.8",
#      build_file = path_prefix + "zlib.BUILD",
# )

# run parsey_api with exported model
$ ./bazel-bin/tensorflow_serving/example/parsey_api --port=9000 ../api/parsey_model
  • node client
$ cd ../api/parsey_client
$ cp index_org.js index.js
$ npm install

# if you have a trouble, check node / npm / grpc
$ node --version
$ npm --version
# you need to install grpc for node
$ npm install grpc

# send sentences to parsey_api server
$ node index.js
{
  "result": [
    {
      "docid": "-:0",
      "text": "This is the first sentence",
      "token": [
        {
          "word": "This",
          "start": 0,
          "end": 3,
          "head": 4,
          "tag": "DT",
          "category": "DET",
          "label": "nsubj",
          "break_level": "SPACE_BREAK"
        },
...
}
  • python client
# you need to install gRPC properly( https://tensorflow.github.io/serving/setup )
# if you have a trouble, see https://github.com/dsindex/tensorflow#tensorflow-serving

# download protobuf_json.py for converting protobuf to json
$ cd ../..
$ git clone https://github.com/dpp-name/protobuf-json.git
$ cp protobuf-json/protobuf_json.py serving/tensorflow_serving/example/

$ cd serving

# create softlink for `parsey_api.proto`
$ ln -s ./tf_models/syntaxnet/syntaxnet syntaxnet

# generate 'parsey_api_pb2.py'
$ which grpc_python_plugin
# if this returns nothing, gRPC was not properly installed. see https://github.com/tensorflow/serving/issues/42
$ protoc -I ./  --python_out=. --grpc_out=. --plugin=protoc-gen-grpc=`which grpc_python_plugin` ./tensorflow_serving/example/parsey_api.proto

# copy parsey_client.py to serving/tensorflow_serving/example
$ cp ../api/parsey_client.py tensorflow_serving/example

# build it
$ bazel --output_user_root=bazel_root build --nocheck_visibility -c opt -s //tensorflow_serving/example:parsey_client --genrule_strategy=standalone --spawn_strategy=standalone --verbose_failures
$ ls bazel-bin/tensorflow_serving/example/parsey_client

# run
$ bazel-bin/tensorflow_serving/example/parsey_client --server=localhost:9000
23:52 $ bazel-bin/tensorflow_serving/example/parsey_client --server=localhost:9000
D0728 23:52:50.764804093   31201 ev_posix.c:101]             Using polling engine: poll
this is a first sentence.
result {
  docid: "-:0"
  text: "this is a first sentence ."
  token {
    word: "this"
    start: 0
    end: 3
    head: 4
    tag: "DT"
    category: "DET"
    label: "nsubj"
  }
  token {
    word: "is"
    start: 4
    end: 5
    head: 4
    tag: "VBZ"
    category: "VERB"
    label: "cop"
  }
...
}
  • export model
# copy parsey_mcparseface.py to serving/tensorflow_serving/example
$ cp ../api/parsey_mcparseface.py tensorflow_serving/example
# build it
$ bazel --output_user_root=bazel_root build --nocheck_visibility -c opt -s //tensorflow_serving/example:parsey_mcparseface --genrule_strategy=standalone --spawn_strategy=standalone --verbose_failures
$ ls bazel-bin/tensorflow_serving/example/parsey_mcparseface

# run
# this will read model from --model_dir and export to --export_path directory
$ bazel-bin/tensorflow_serving/example/parsey_mcparseface --model_dir=syntaxnet/models/parsey_mcparseface --export_path=exported

# modify all path in exported/00000001/assets/context.pbtxt
# for example, 
# from
# input {
#  name: "tag-map"
#  Part {
#    file_pattern: "syntaxnet/models/parsey_mcparseface/tag-map"
#  }
# }
# to
# input {
#  name: "tag-map"
#  Part {
#    file_pattern: "tag-map"
#  }
# }

# run parsey_api with exported model
$ ./bazel-bin/tensorflow_serving/example/parsey_api --port=9000 exported/00000001


# if you want to export a model trained by the same version of syntaxnet(==tf_model/syntaxnet), 
# copy trained models to current directory and set proper path in models/context.pbtxt
# ex) file_pattern: 'OUTPATH/label-map' -> file_pattern: '/path/to/label-map'
$ cat models/context.pbtxt.template | sed "s=OUTPATH=/path/to=" > models/context.pbtxt
$ bazel-bin/tensorflow_serving/example/parsey_mcparseface --model_dir=models --export_path=exported

# modify all path in exported/00000001/assets/context.pbtxt since those paths are not neccessary.
# for example, 
# from
# input {
#  name: "tag-map"
#  Part {
#    file_pattern: "syntaxnet/models/parsey_mcparseface/tag-map"
#  }
# }
# to
# input {
#  name: "tag-map"
#  Part {
#    file_pattern: "tag-map"
#  }
# }

# run parsey_api with exported model
$ ./bazel-bin/tensorflow_serving/example/parsey_api --port=9000 exported/00000001
  • what about parsing only case? especially when you trained Korean parser.
# the parsey_api server can handle conll format.
# so, just export model and use it

# export parsing model only
# replace serving/tensorflow_serving/example/parsey_mcparseface.py with parsey_sejong.py
$ cp ../api/parsey_sejong.py tensorflow_serving/example/parsey_mcparseface.py
# build it
$ bazel --output_user_root=bazel_root build --nocheck_visibility -c opt -s //tensorflow_serving/example:parsey_mcparseface --genrule_strategy=standalone --spawn_strategy=standalone --verbose_failures
$ ls bazel-bin/tensorflow_serving/example/parsey_mcparseface

# copy trained models_sejong to current directory and set proper path in models_sejong/context.pbtxt
# ex) file_pattern: 'OUTPATH/label-map' -> file_pattern: '/path/to/label-map'
$ cat models_sejong/context.pbtxt.template | sed "s=OUTPATH=/path/to=" > models_sejong/context.pbtxt

# run
# this will read model from --model_dir and export to --export_path directory
$ bazel-bin/tensorflow_serving/example/parsey_mcparseface --model_dir=models_sejong --export_path=exported_sejong

# modify all path in exported_sejong/00000001/assets/context.pbtxt since those paths are not neccessary.
# for example, 
# from
# input {
#  name: "tag-map"
#  Part {
#    file_pattern: "syntaxnet/models/parsey_mcparseface/tag-map"
#  }
# }
# to
# input {
#  name: "tag-map"
#  Part {
#    file_pattern: "tag-map"
#  }
# }

# run parsey_api with exported model
$ ./bazel-bin/tensorflow_serving/example/parsey_api --port=9000 exported_sejong/00000001

# node client
# send conll format to parsey_api server
$ cd ../api/parsey_client
$ cp index_sejong.js index.js
$ node index.js | more
{
  "result": [
    {
      "docid": "-:0",
      "text": "내 가 집 에 가 ㄴ다 .",
      "token": [
        {
          "word": "내",
          "start": 0,
          "end": 2,
          "head": 1,
          "tag": "NP",
          "category": "NP",
          "label": "MOD",
          "break_level": "SPACE_BREAK"
        },
        {
          "word": "가",
          "start": 4,
          "end": 6,
          "head": 4,
          "tag": "JKS",
          "category": "JKS",
          "label": "NP_SBJ",
          "break_level": "SPACE_BREAK"
        },
...
}

cd ../..
cd serving

# python client
# replace serving/tensorflow_serving/example/parsey_client.py with parsey_sejong_client.py
$ cp ../api/parsey_sejong_client.py tensorflow_serving/example/parsey_client.py

# parsey_sejong_client.py import konlpy
# so, you need to install konlpy( http://konlpy.org/ko/v0.4.3/install/ )

# build it
$ bazel --output_user_root=bazel_root build --nocheck_visibility -c opt -s //tensorflow_serving/example:parsey_client --genrule_strategy=standalone --spawn_strategy=standalone --verbose_failures
$ ls bazel-bin/tensorflow_serving/example/parsey_client

# run
$ bazel-bin/tensorflow_serving/example/parsey_client --server=localhost:9000
나는 학교에 간다
nput :  나는 학교에 간다
Parsing :
{"result": [{"text": "나 는 학교 에 가 ㄴ다", "token": [{"category": "NP", "head": 1, "end": 2, "label": "MOD", "start": 0, "tag": "NP", "word": "나"}, {"category": "JX", "head": 4, "end": 6, "label": "NP_SBJ", "start": 4, "tag": "JX", "word": "는"}, {"category": "NNG", "head": 3, "end": 13, "label": "MOD", "start": 8, "tag": "NNG", "word": "학교"}, {"category": "JKB", "head": 4, "end": 17, "label": "NP_AJT", "start": 15, "tag": "JKB", "word": "에"}, {"category": "VV", "head": 5, "end": 21, "label": "MOD", "start": 19, "tag": "VV", "word": "가"}, {"category": "EC", "end": 28, "label": "ROOT", "start": 23, "tag": "EC", "word": "ㄴ다"}], "docid": "-:0"}]}