MLPQNA model

[cavuoti]

No description provided.

[EiffL]

@cavuoti Thanks for this! Could you add a quick description of the method in you PR? I've looked through the code and I see you are using neural networks :-)

Also, would you be available on Tuesday to present a few slides on you results at the DESC collaboration meeting? There is a longer message about this on the LSSTC Slack #desc-2x3pt channel

[cavuoti]

Hi @EiffL,
I am sorry, I sent a couple of links to Joe about the method but the smartest thing would be to post something here.

Anyway, the method is indeed a neural network with on top a standard scaler, in particular we choose a configuration that we call MLPQNA and that we used several times in the past (mostly for photometric redshift, see for instance https://arxiv.org/abs/1206.0876) but we used it also for classification (see for instance this: https://arxiv.org/abs/1504.03857)
We made a search for the best parameters and the one that are hardcoded are the ones that we think are the best.

if you think it is usefull I can repeat a description here but the one given in the first link I think should be fine.
For what concerns the presentation we can for sure add a couple of slides for tomorrow.

[EiffL]

This is perfect! Thanks so much!