/ds_projects

My personal projects in Data Science

Primary LanguageJupyter Notebook

ds_projects

My personal projects in Data Science

ds_projects/hf_mlp_flask_docker/

The neural network model to predict the heat of formation of a compound based on its chemical formula. The accuracy (MAE) is about 43 meV/atom, reaching the so-called Chemical Accuracy, and hence allowing for realistic applications without complicated atomic simulations.

A simple Web API with Flask and a Dockerfile are included for deployment on AWS, etc. You can use the model via the webpage, as well as cURL if the command line way is preferred.

$ python app.py
$ curl -X POST 0.0.0.0:5000/curl -H 'Content-Type: application/json' -d '"Sn1O1 Sn1O2 Sn2O3 Zn1O1 Zn1Sn1O2"'
Sn1O1 -1.570 ev/atom
Sn1O2 -1.859 ev/atom
Sn2O3 -1.483 ev/atom
Zn1O1 -1.606 ev/atom
Zn1Sn1O2 -1.450 ev/atom

To test the docker, first build and run the image:
$ docker build -t app-hf .
$ docker run -p 5000:5000 app-hf
and then use the curl command line:
$ curl -X POST 0.0.0.0:5000/curl -H 'Content-Type: application/json' -d '"Sn1O1 Sn1O2 Sn2O3 Zn1O1 Zn1Sn1O2"'

More is coming soon