nmr_timeseries_model

Code for "Using Machine Learning to Improve the Hard Modeling of NMR Time Series" paper

Requirements

MATLAB-Code: Matlab 2021b or later, Python 3.8.10 or later

Python-Code: Python 3.8.10 or later, torch, numpy, scipy, matplotlib

Modeling data of NMR time series, quick start

Getting started: Run main.m to get a quick glance at the program. If you want to use machine learning, define the correct path (in main.m) of the virtual environment which knows pytorch and numpy.

Loading data sets: Requires a phase and baseline-corrected data set as a .mat file. The file should contain 3 different variables.

• The data, called d or D
• The frequency axis, called x or X
• The time axis, called t or T.

Then, create new .m file for loading data, which can be copied and pasted from existing files in data/ and add it to file section in main.m.

Set hyperparameters in options section in main.m (not necessary, but incorrect definition of hyperparamters leads to incorrect results).

Then, run main in desired variant.

Results can be saved, if opt.isSaveFile = 1;

Training CNNs

New NNs can be trained using the designated Python-Code for training neural networks (Training_Nets_Python/main_pytorch.py). Keep in mind that each new network should predict different numbers of peaks. Hyperparameter are setup for testing and should be optimized.

Once the networks are fully trained, save them under python/pytorch/MATLAB_net_[n]p_SD.pt, where [n] is the number of peaks. DON'T store unused newtorks in this folder!