tfcgo was born out of necessity. Tensorflow go bindings are slow to progress. Relying on protobufs to generate wrapper code is dependant on the growth of the c/c++ public api - and the underlying framework has much more to offer! This library will eventually do the hard work bridging the gap between go and the Tensorflow c++ framework. Training a TF machine learning algorithm is possible in go. Please note contributions in this repo will be submitted to the main tensorflow repo.

Install Docker

git clone this repo: git clone https://github.com/ctava/tfcgo

Run the following commands:
cd tfcgo
docker build -t ctava/tfcgo .
To run with shared volume, you can add /go to shared paths from Docker -> Preferences... -> File Sharing.
docker run -v /go/src/github.com/ctava/tfcgo:/go/src/github.com/ctava/tfcgo -it -p 8888:8888 ctava/tfcgo:latest

and your in. You now have tensorflow + golang + tfcgo available.

Run the following commands:
docker pull ctava/tfcgo:latest
docker run -it --security-opt=seccomp:unconfined -p 8888:8888 ctava/tfcgo:latest

--security-opt for the purposes of using delve

and your in. You now have tensorflow + golang + tfcgo available.

Run the following commands:
go run versioncheck.go

You should see the versions of golang, tensorflow and tfcgo as specified in the docker file.

go run listops.go

You should see a list of all of the ops available in tensorflow

Training

go run linear.go

Multi-Class Classification

go run iris.go (coming soon)

gopherdata