Provide instructions for Bazel build about models HOT 4 CLOSED

I hope you provide a simpler method for at least flowers, that does not include complete bazel build since it took 40+ mins only to fail because I did not execute configure and even then it failed again with some other error while building tensorflow. I already had a working tensorflow installation inside a virtualenv.

This was the error, it belongs to tf repo. But at this point I am just going to decode what exactly Bazel build does when its building flowers_train. And going to manually reconstruct those steps rather than dealing with building Tensorflow when I have it already installed.

3 warnings generated.
ERROR: /private/var/tmp/_bazel_aub3/cc8ffd9a09258bb3405e23c56942c269/external/tf/tensorflow/python/BUILD:915:1: SWIGing external/tf/tensorflow/python/tensorflow.i failed: swig failed: error executing command bazel-out/host/bin/external/tf/tensorflow/swig -c++ -python -module pywrap_tensorflow -lexternal/tf/tensorflow/python/client/device_lib.i -lexternal/tf/tensorflow/python/client/events_writer.i ... (remaining 190 argument(s) skipped): com.google.devtools.build.lib.shell.BadExitStatusException: Process exited with status 127.
bazel-out/host/bin/external/tf/tensorflow/swig: line 17: swig: command not found
Target //inception:flowers_train failed to build

After installing swig via brew it finally seems to be building, though not finished yet.

from models.

I finally ended up editing WORKSPACE and BUILD, removed all references to tensorflow.
This ensured that bazel did not build it. Since everywhere tensorflow is imported via python, there is no need to bundle it as bazel dependancy.

Bazel wont use python present inside a virtualenv....

from models.

Sorry, I will fix this. There should be a tensorflow submodule. We may still consider requiring TF to be installed seprately, but this is a bug.

from models.

We removed the submodule. I'll close this, please reopen or make a new one if you see other issues.

Thanks!

from models.