Skip to content
Complementary code for the Targeted Dropout paper
Python
Branch: master
Clone or download
Latest commit 87cf2a0 Jun 17, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data Upstream push (#3) Dec 2, 2018
hparams Upstream push (#3) Dec 2, 2018
models Upstream push (#3) Dec 2, 2018
scripts fix eval script Jun 16, 2019
training Upstream push (#3) Dec 2, 2018
.gitignore initial commit Oct 17, 2018
.travis.yml Upstream push (#3) Dec 2, 2018
README.md initial commit Oct 17, 2018
__init__.py initial commit Oct 17, 2018
requirements.txt Upstream push (#3) Dec 2, 2018
train.py Upstream push (#3) Dec 2, 2018

README.md

Targeted Dropout

Aidan N. Gomez, Ivan Zhang, Kevin Swersky, Yarin Gal, and Geoffrey E. Hinton

Table of Contents

Requirements

  • Python 3
  • Tensorflow 1.8

Quick Start

  1. Train a model: python -m TD.train --hparams=resnet_default
  2. Prune that model: python -m TD.scripts.prune.eval --hparams=resnet_default --prune_percent 0.0,0.25,0.5,0.75,0.95

Flags

  • --env: one of local, gcp (GPU instances), or tpu (TPU instances). Feel free to add more if necessary.
  • --hparams: the hparam set you want to run.
  • --hparam_override: manually specify hparams to be overridden (e.g --hparam_override 'drop_rate=0.66')
You can’t perform that action at this time.