Blinding?

[chihway]

From Troxel: "Does DESC have a blinding policy that will guide these measurements/constraints? We are working with state-of-the-art data now, and the results will be meaningful and should be properly blinded. "

Let's discuss where and when and how we want to blind.

[EiffL]

That's a very good point... but that will require a fair amount of more thinking than I was expecting for this project. I'm all for blinding though, and I guess the sooner we start putting the infrastructure and method in place the better. @joezuntz do you have thoughts on where when and how to incorporate some blinding strategies in TXPipe?

[joezuntz]

This is a very good point. We could incorporate it either in the 2pt estimation functions themselves, or add a blinding stage following on from them I guess. In terms of methodology - the one presented in Muir et al recently is very nice as it can be applied to any statistic we calculate.

[chihway]

Let's default to apply the Muir et al. blinding (which we'd need to incorporate into TXPipe) if we don't have a clear, better alternative.

[chihway]

Starting a branch in TXPipe to incorporate this blinding. To incorporate Muir et al. method we would need to interface with CCL. Let's plan on doing this from the start. In order to not slow down the development too much, I'll have two options for blinding: 1) multiplying the data vector by a random number before saving it to a sacc file 2) full Muir et al implementation. We can use 1) right away while we work on 2).

[slosar]

So, the Muir recipe is additive, is this the right approach or do you want to make them multiplicative (i.e. performing Muir et al in log of measrurements)? I'm actually not sure, it does seem like the arguments for either (also dependening on what do you think you leading systematics are).

I think implementing this could be a nice hack day project.

[slosar]

I have now read Muir et al again more carefully and I agree with them that the shifts should be additive...

[chihway]

A possible way to do this:

• call firecrown compute twice, with different parameters
• generate delta data vector from the two sacc files above
• add to txpipe two-point

Alternatively, call CCL see: https://github.com/LSSTDESC/TXPipe/blob/master/txpipe/plotting/correlations.py

[chihway]

This is in TXPipe now. https://github.com/LSSTDESC/TXPipe/pull/63. Closing.