Predict the lowest energy level of a one-dimensional ultracold atom subject to an external optical speckle field (disorder). Our aim is to solve the Schrödinger equation and thus find the the ground state energy.
We want to reproduce the results achieved in Scientific reports 9.1 (2019): 1-12., but using a scalable version of the network, as did in Physical Review E 102.3 (2020): 033301.
Install all the requirements using pip install -r requirements.txt, we recommend to create an isolated environment before.
The input data must be organized as follow:
data
├── train_data_L14.npz
├── test_data_L14.npz
├── train_data_L28.npz
├── test_dataL_28.npz
└── ...
They are not present in the git repo, but they are available in my own Drive.
To train with the simplest MLP model one has to run
python main.py --train --data_dir data/train_dataL14.npz --input_size 15.
Several commands are avialable to switch model, change MLP hidden structure, etc. Just run python main.py --help.