I have installed Tensorflow, bazel both latest version.
To train a model from scratch I have to run the following command on this link https://github.com/tensorflow/models:
bazel-bin/inception/imagenet_train --num_gpus=1 --batch_size=32 --train_dir=/tmp/imagenet_train --data_dir=/tmp/imagenet_data
It gives an error
bazel-bin/inception/image_train: No such file or directory
bazel-bin seems to be file and not a directory.
Further, if try to go to /models/inception/inception path and try to run the imagenet_train.py file it throws an error:
command not found error
I have no idea why isn't it working. I have followed each and every step. This is bugging me for a long time now.
Original answer:
You have to build
imagenet_trainfirst, what is the output when you runbazel build //inception:imagenet_train?
bazel-binis a symbolic link to a directory.
Based on your comment below (~/models#), it looks like you're running Bazel in the wrong directory. You have to cd to the inception/ directory before running bazel:
cd inception
/opt/DL/bazel/bin/bazel build //inception:imagenet_train
Tensorflow chose a very odd project structure: models/ is a project, but every subdirectory of models is also its own project. I'm not sure why they did this, but you have to build the subprojects (like inception) in their own directories, not the top-level directory.
//inception:imagenet_train is called a target. Everything before the : tells you where the target is defined (the BUILD file in the inception/ directory). Tensorflow made this even more confusing by putting everything in a subdirectory named the same way as its project (e.g., this target is defined in ~/models/inception/inception/BUILD).
imagenet_train is the identifier for the target, you can see its definition here.
See Bazel's Getting Started docs for a more detailed explanation of what a target is.