https://github.com/datainq/go-mnist-client

TensorFlow serving mnist example client in Go.
https://github.com/datainq/go-mnist-client

go golang grpc tensorflow tensorflow-examples tensorflow-tutorials

Last synced: 4 months ago
JSON representation

TensorFlow serving mnist example client in Go.

• Host: GitHub
• URL: https://github.com/datainq/go-mnist-client
• Owner: datainq
• License: apache-2.0
• Created: 2018-01-16T18:18:11.000Z (about 7 years ago)
• Default Branch: master
• Last Pushed: 2018-01-22T06:35:34.000Z (about 7 years ago)
• Last Synced: 2024-09-28T10:04:11.465Z (4 months ago)
• Topics: go, golang, grpc, tensorflow, tensorflow-examples, tensorflow-tutorials
• Language: Go
• Size: 103 KB
• Stars: 17
• Watchers: 5
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# go-mnist-client

An example of tensorflow_serving client in Go. 

It demonstrates how to connect to the `tensorflow_model_server` launched

as described in 

[Serving a TensorFlow Model](https://www.tensorflow.org/serving/serving_basic) on

TensorFlow documentation webpage.

The main problem with tensorflow_serving is a lack of Go package ready to 

import and run.

This example is based on:

[github.com/tobegit3hub/tensorflow_template_application](https://github.com/tobegit3hub/tensorflow_template_application)

## Setup fake workspace through a`$GOPATH` or copy files

For a convenience, I provide the generated files under the `gopath` 

subdirectory. You can copy files directly to your `$GOPATH`.

```bash

cp gopath/src $GOPATH

```

## Generating from sources

For a purpose of generating proto we will use a `docker` container `grpc/go` with 

`protoc`, `go` and [`gRPC`](https://grpc.io/) protoc extension installed.

Remember that a version of TensorFlow Serving, TensorFlow C++ library and

and Go TensorFlow library versions should match.

```

go get -u golang.org/x/net/context

go get -u github.com/golang/protobuf/{proto,protoc-gen-go}

go get -u google.golang.org/grpc

go get -u github.com/tensorflow/tensorflow/tensorflow/go

go get -u go.uber.org/ratelimit

```

Let's start a docker container mounting a fake gopath:

```bash

docker run --rm -it -v $PWD/gopath:/gopath:ro -v $PWD/gen:/gen grpc/go /bin/bash

```

and copy paste:

```

export GOPATH=/gopath

git clone https://github.com/tensorflow/serving

git clone https://github.com/tensorflow/tensorflow

IN=$PWD

OUT=/gen

protoc -I=$IN/serving -I $IN/tensorflow \

    --go_out=plugins=grpc:$OUT \

    $IN/serving/tensorflow_serving/apis/*.proto

IN=$PWD/tensorflow

protoc -I=$IN --go_out=plugins=grpc:$OUT \

    $IN/tensorflow/core/framework/*.proto

    

protoc -I=$IN --go_out=plugins=grpc:$OUT \

    $IN/tensorflow/core/protobuf/{saver,meta_graph}.proto

    

protoc -I=$IN --go_out=plugins=grpc:$OUT \

    $IN/tensorflow/core/example/*.proto

```

In most cases we have to fix permissions:

```bash

sudo chown -R $USER.$USER gen

```

Most of the tensorflow_serving and tensorflow is not prepared to be

imported in go directly, therefore we must **fix incorrect import paths**:

``` 

grep -rl '"tensorflow/' --include \*.go gen | 

xargs sed -i 's/"tensorflow\//"github.com\/tensorflow\/tensorflow\/tensorflow\/go\//g'

```

And now we need to copy files into the GOPATH:

```bash

mkdir -p $GOPATH/src/github.com/tensorflow/serving

cp -R gen/tensorflow_serving $GOPATH/src/github.com/tensorflow/serving/

mkdir -p $GOPATH/src/github.com/tensorflow/tensorflow/tensorflow/go/

cp -R gen/tensorflow/core $GOPATH/src/github.com/tensorflow/tensorflow/tensorflow/go/

```