AISC

Artificial Intelligence for SuperConductors

This repository is a data-science project related to superconduttivity phenomena. The project tackles both classification and regression problems and, making use of the Deep Set architecture, provide a new characterization of chemical elements to help discovery and understood superconductors.

Set up

To assist reproducibility and ease of use we provide a dockerfile to build a container with the necessary depedencies.

 sudo docker image build -t aisc:latest .
 sudo docker container run -it --rm -p 8888:8888  --volume $(pwd):/AISC aisc

The last command will open a terminal within which we are able to reproduce the experiments conducted.

Project Structure

The folder structure is:

AISC
└── project_aisc
    ├── config
    ├── data
    │   ├── external <- hosono.csv
    │   ├── processed
    │   └── raw <- supercon.csv & garbagein.csv
    ├── models
    ├── notebooks
    ├── reports
    └── src
        ├── data
        ├── features
        ├── laboratory
        ├── model
        └── utils

In config there are config files (yalm) to customize models and training process.
Notebooks contain notebooks to analyze the data we had and we produced during the experiments.
The code used to produce the results is stored into laboratory;

Run Experiments

We can reproduce 3 experiments:

#We can run the experiments from any folder
#First move in the project_aisc folder
cd project_aisc

#Train and test the model
# Pipeline is managed by dvc software. If no dvc file is present execute
dvc init

# Reproduce pipeline
dev repro

# DVC skip stages if they are alredy executed. To force again execution of the entire pipeline 
dvc repro -f 

#Produce latent features and inspect them
python src/laboratory/latent_space.py

#Compare difference features strategy
python src/laboratory/compare_features_strategy.py

With the appropriate flags is possible to use customized model and save the results.

Analyze Experiments

We can both do eda on SuperCon data or on experiment results

#Move inside notebook
cd notebook/

#Open the browser and pick up the desired exploration
jupyter-nbclassic

#Change kernel to python369 from kernel tab