neural networks to generate parameters

ergodicio / tsadar

The Thomson scattering diagnostic offers a method by which to infer plasma parameters such as n_e, T_e. This approach uses the form factor equations to estimate those plasma parameters by fitting to observed Thomson scattering spectra.

2 stars 2 forks source link

neural networks to generate parameters #19

Closed joglekara closed 1 year ago

joglekara commented 1 year ago

implements

neural network parameter generators. the NN takes in the spectrum and provides the parameters that then go into the TS code
batch-wise training
csv outputs
plotting best and worst

can fit 200 slices in about 10 minutes, hyperparameter search pending

joglekara commented 1 year ago

This is now passing tests.

We still need to make the network bigger or smaller depending on how big of a dataset we're working with. So in practical use, it may require a few tries before we're able to fit a dataset

joglekara commented 1 year ago

this is now working with parameter learning with SGD (stochastic gradient descent)