Skip to content

alshedivat/adios

master
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
mbc
 
 
 
 
 
 
 
 
 
 
 
 

ADIOS: Architectures Deep In Output Space

ADIOS is implemented as a thin wrapper around Keras' Graph model (i.e., multiple-input multiple-output deep architecture) by adding the adaptive thresholding functionality as described in the paper. adios.utils.assemble.assemble helper function provides and handy way to construct ADIOS and MLP models from config dictionaries. Configs can be generated from templates using adios.utils.jobmab.gen_configurations. Examples of templates are given in configs/ folder in YAML format. Additionally, we provide utility functions for hyperparameter or architecture search using Jobman.

All example scripts are given in scripts/.

Note: keras.models.Graph is no longer supported starting from keras-v1.0 as of April, 2016. The current version of ADIOS uses the legacy code, keras.legacy.models.Graph.

Requirements

  • NumPy
  • pyyaml
  • Theano
  • keras>=1.0
  • scikit-learn

The requirements can be installed via pip as follows:

$ pip install -r requirements.txt

Optional (needed only for using Jobman):

Installation

To use the code, we recommend installing it as Python package in the development mode as follows:

$ python setup.py develop [--user]

The --user flag (optional) will install the package for a given user only.

Other implementations

Citation

@inproceedings{cisse2016adios,
  title={ADIOS: Architectures Deep In Output Space},
  author={Cisse, Moustapha and Al-Shedivat, Maruan and Bengio, Samy},
  booktitle={Proceedings of The 33rd International Conference on Machine Learning},
  pages={2770-–2779},
  year={2016}
}

License

MIT (for details, please refer to LICENSE)

Copyright (c) 2016-2018 Moustapha Cisse, Maruan Al-Shedivat, Samy Bengio