GDHF demo

[JulienVig]

Implements a DISCO demo for GDHF with the help of @Mastel22 and @tharvik. The demo consists in binary dog cuteness classification

Running the demo locally

Download the dataset zip
Setup the project

git clone git@github.com:epfml/disco.git
cd disco
git switch gdhf # go to the gdhf branch
npm ci && npm -ws run build # install and build the project

3. In the meantime, unzip the dataset into the datasets/ folder
4. When the project is built, start the server, the webapp and the CLI in different tabs

npm -w server start # 1st terminal

npm -w webapp start # 2nd terminal

npm -w cli start # 3rd terminal

5. Go to the webapp in your browser: http://localhost:8081/
6. Go to the GDHF task
7. Connect the images in the 0th split: datasets/tinder_dog/0 by CSV (not by group)
8. Click on Train
9. When the training is done click on next to test the model
10. You can use the data in the test folder

Running the demo on discolab.ai

At step 4, add --host option when starting the CLI:

npm -w cli start -- --host https://disco-zbkj3i466a-oa.a.run.app/

At step 5, go to discolab.ai

demo.mov

[tharvik]

all ready for the demo, let's go!
very few non-blocking changes