A main loop to use Tensorflow with custom data not in the protobuf format. This code is optimized to work coupled with the dataset loaders, a framework to load and preprocess your data with parallel threads.

If you use this code, please cite:

• [1] Francesco Visin - Main loop TF: a main loop for Tensorflow and custom data (BibTeX)

If you use the Dataset loaders, please also cite:

• [1] Francesco Visin, Adriana Romero - Dataset loaders: a python library to load and preprocess datasets (BibTeX)

To use the main loop, simply create a file with this code and call it.

if __name__ == '__main__':
    import sys

    from main_loop_tf.main import run

    from model import build_model

    argv = sys.argv
    # Here you can potentially put some constant params you want to pass to the
    # main loop
    run(argv, build_model)

To add some model specific parameters to the list of parameters, just specify them in your model file with the usual gflags syntax:

    import gflags

    gflags.DEFINE_integer('my_custom_param', 0, 'This sets up a custom param')

    def build_model():
        cfg = gflags.cfg
        print('This is your param value {}'.format(cfg.my_custom_param))
        pass

You can find a list of DEFINE* methods here

You can also flag some options as required with:

    gflags.mark_flag_as_required('required1')
    gflags.mark_flags_as_required(['required1', 'required2])

To add list of lists (e.g. [[10, 10], [20, 20]]) you can use the gflags extensions in gflags_ext:

import gflags
import sys
from main_loop_tf import gflags_ext

gflags_ext.DEFINE_intlist('a', [[10, 10], [20, 20]], 'A list of ints')

See gflags_ext for the other DEFINEs.

The models will be saved in: <checkpoints_basedir>(/<suite_name>)/<model_name>(_model_suffix)

and restored from: <checkpoints_basedir>(/<restore_suite>)/<restore_model>

• model_name and restore_model: default to the hash of the hyperparameters if not specified
• suite_name, restore_suite and model_suffix: are ignored if not specified
• checkpoints_path is set to <checkpoints_basedir>(/<suite_name>)

• The code is provided as is, please expect minimal-to-none support on it.
• This code is provided for research purposes only. Although we tried our best to test it, the code might be bugged or unstable. Use it at your own risk!
• PRs welcome!! Feel free to contribute to the code with a PR if you find bugs or want to improve the existing code!

The initial version of the "main loop" has been inspired by a simpler pipeline developed by Francesco Lattari (a.k.a. "er latta") and Marco Ciccone, based on the Tensorflow multi-GPU example.

I am thankful to both for being early adopters and enthusiastic supporters of this framework, for their code contributions, but most of all for the outstanding number of bugs and issues they found and raised.

You guys really made my life miserable, thank you!! :)

THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.