cogsci-2019-concept-learning

Code for our 2019 CogSci paper "Learning deep taxonomic priors for concept learning from few positive examples." To cite the work that this code is associated with, use:

@inproceedings{grant2019learning,
  title={Learning deep taxonomic priors for concept learning from few positive examples},
  author={Grant, Erin and Peterson, Joshua C and Griffiths, Thomas L},
  booktitle={Proceedings of the Annual Conference of the Cognitive Science Society},
  year={2019}
}

tl;dr

1. Install the package. (Remember to conda activate cogsci-2019-concept-learning if necessary.)

2. Run the following commands to set up NLTK:

python -c "import nltk; nltk.download('wordnet'); nltk.download('omw-1.4')"

3. TODO(eringrant): Document how to get ImageNet images by synset.

4. A training and evaluation run on the human data can then be run via:

scripts/run_human_comp.sh /tmp /tmp PATH_TO_IMAGENET

Installation

Option: Conda install

To install via Conda, do:

git clone git@github.com:eringrant/cogsci-2019-concept-learning.git
cd cogsci-2019-concept-learning
conda env create --file environment.yml

The Conda environment can then be activated via conda activate cogsci-2019-concept-learning.

Option: pip install

To install via pip, do:

git clone git@github.com:eringrant/cogsci-2019-concept-learning.git
cd cogsci-2019-concept-learning
pip install -e .