act_nilc_noise_070123

[AdriJD]

Added act_dr6v4_nilc070123 dataset and products to sofind that can be used to draw mnms noise simulations of the NILC Compton y maps.

Some notes

• Each of 6 mnms models has been estimated from 10 input sims containing identically distributed noise. This is not a mode that mnms currently supports, so the model generation has been handled by a bespoke script. As a result, the noise models configs included here can only be used to draw simulations, they cannot be used to generate new models. No input maps are included in the release for this reason as well.
• The input simulations from which the mnms models were generated came in two forms: wavelet-based sims and tiled-based sims. As a result, there are two separate map products (act_dr6v4_nilc070123_tiled.yaml and act_dr6v4_nilc070123_wav.yaml) The tile and wav models included in act_dr6v4_nilc070123.yaml correspond to the tile and wavelet input maps, respectively.

[zatkins2]

Thanks. I'm just a little confused by what is happening here.

You say there are no input maps included, but there are maps products that are added, including with system paths? Is that just because those fields/products need to be supplied to make the code work?

And as far as the noise models go, you're saying that the noise model that produces e.g. tile sims was built off of input tile sims that passed through the ILC pipeline? Is there any reason why someone couldn't make wavelet sims that are built off the input tile sims? And did you check that everything works for drawing sims in spite of most fields being null?

Minor question, but what is the difference between e.g. tile and tile_planck noise models? Just the mask_obs_name?

[AdriJD]

The map products are there because the tile models need the inverse variance maps. But thinking about it, the wavelet models don't need those. So I could remove the act_dr6v4_nilc070123_wav.yaml map product.

We could make a wavelet model based on the tile sims, or vice versa. But I think having two types of simulations is enough. I don't really see the point in mixing the wavelet and tile methods for this application.

I did check that drawing of sims works. The null fields in the configs would be not-null if we would allow generation of models with these configs. For drawing sims, this works as far as I have tested.

Difference between tile and tile_planck is just the mask. For wav and wav_planck it also the w_lmax_j parameter

[zatkins2]

Ok -- how do you get the ivar map for the tiles? This would be the ivar map for the ILC reconstruction, right?

[AdriJD]

It is estimated from the original NILC sims. It's indeed the ivar map for the NILC maps, it looks different than the ACT ivar maps.