Skip to content
Branch: master
Find file History
yesmung Modify flag name for the checkpoint path
change flag name to checkpoint_dir according to the variable name
used by the checkpoint_utils within tensorflow python framework.

The important point is that when run the run_eval script, an error
occurs due to the different flag name.

Signed-off-by: MyungSung Kwak <yesmung@gmail.com>
Latest commit 2949cfd Aug 13, 2018
Permalink
Type Name Latest commit message Commit time
..
Failed to load latest commit information.
architectures First commit of learning_unsupervised_learning Apr 3, 2018
datasets First commit of learning_unsupervised_learning Apr 3, 2018
meta_objective First commit of learning_unsupervised_learning Apr 3, 2018
.gitignore
README.md First commit of learning_unsupervised_learning Apr 3, 2018
__init__.py First commit of learning_unsupervised_learning Apr 3, 2018
evaluation.py First commit of learning_unsupervised_learning Apr 3, 2018
optimizers.py First commit of learning_unsupervised_learning Apr 3, 2018
run_eval.py Modify flag name for the checkpoint path Aug 14, 2018
summary_utils.py First commit of learning_unsupervised_learning Apr 3, 2018
utils.py First commit of learning_unsupervised_learning Apr 3, 2018
variable_replace.py First commit of learning_unsupervised_learning Apr 3, 2018

README.md

Learning Unsupervised Learning Rules

This repository contains code and weights for the learned update rule presented in "Learning Unsupervised Learning Rules." At this time, this code can not meta-train the update rule.

Structure

run_eval.py contains the main training loop. This constructs an op that runs one iteration of the learned update rule and assigns the results to variables. Additionally, it loads the weights from our pre-trained model.

The base model and the update rule architecture definition can be found in architectures/more_local_weight_update.py. For a complete description of the model, see our paper.

Dependencies

absl, tensorflow, sonnet

Usage

First, download the pre-trained optimizer model weights and extract it.

# move to the folder above this folder
cd path_to/research/learning_unsupervised_learning/../

# launch the eval script
python -m learning_unsupervised_learning.run_eval \
--train_log_dir="/tmp/learning_unsupervised_learning" \
--checkpoint_dir="/path/to/downloaded/model/tf_graph_data.ckpt"

Contact

Luke Metz, Niru Maheswaranathan, Github: @lukemetz, @nirum. Email: {lmetz, nirum}@google.com

You can’t perform that action at this time.