depiction

A collection of tools and resources to interpret deep learning models in a framework-independent fashion.

The core of the repo is a package, called depiction, with wrappers around models and methods for interpretable deep learning.

DISCLAIMER: This repo is undergoing a refactoring. For the latest developments (e.g. as shown in ISMB/ECCB 21), please check the other branches, in particular visualizations.

Docker setup

Install docker

Make sure to have a working docker installation. Installation instructions for different operative systems can be found on the website.

Get drugilsberg/depiction image

We built a docker image for depiction containing all models, data and dependencies needed to run the notebooks contained in the repo. Once the docker installation is complete the depiction image can be pulled right away:

docker pull drugilsberg/depiction

NOTE: the image is quite large (~5.5GB) and this step might require sometime.

Run drugilsberg/depiction image

The image can be run to serve jupyter notebooks by typing:

docker run -p 8899:8888 -it drugilsberg/depiction

At this point just connect to http://localhost:8899/tree to run the notebooks and experiment with depiction.

Daemonization

We recommend to run it as a daemon:

docker run -d -p 8899:8888 -it drugilsberg/depiction

maybe mount your local notebooks directory to keep the changes locally

docker run --mount src=`pwd`/notebooks,target=/workspace/notebooks,type=bind -p 8899:8888 -it drugilsberg/depiction

and stopped using the container id:

docker stop <CONTAINER ID>

Development setup

Setup a conda environment

conda env create -f environment.yml

Activate it:

conda activate depiction-env

Install the module:

pip install .