Test 0.2.0

[zacharymartinot]

0.2.0 -- Power-law EoR signal

The power spectrum is recovered by a hera_pspec estimator from a simulated signal visibility

Contributors:

• @zacharymartinot

Some questions

• [ ] All tests passed
• [x] I used/followed the notebook template
• [x] All HERA software has full git commit hash referenced
• [x] All data used is explicitly referenced

Steps Forward

A suggested list of condensed automatic tests arising from this PR, and relevant repos to include them in:

A suggested list of follow-up validation tests which may not already appear on the validation path:

• There is disagreement between the estimate and the input power spectrum that requires further investigation and explanation

[steven-murray]

From telecon discussion:

• @aelanman : residual structure seems to be correlated as new samples are added
• @zacharymartinot : yes the window function drives correlation
• @aelanman : it looks like its not converging quite to zero, but something that oscillates around zero.
• @zacharymartinot : I think it will eventually converge to zero.
• @steven-murray : could you show also the standard error on the mean to check this?
• @zacharymartinot : I do already compute that.
• @steven-murray : would it make sense to show it on the fractional resids?
• @zacharymartinot : yes, I can add that.
• @steven-murray : please put in suggestions on ways forward (low-k discrepancy) into main comment above.

[zacharymartinot]

• Abstract: "signal to measured" --> signal to be measured
  • fixed
• Description: "Last executed" has not been executed
  • fixed
• Summary: apparently \sim is not defined, just use ~
  • fixed
• Summary: explicitly check Lanman 2019 instead of "recalling"
  • removed, not relevent
• Summary: can we say something about the residual distribution here, rather than just % deviation? Like how many sigma are residuals from zero?
  • not necessary for this notebook
• Software: Needs to be filled out properly. Do you have the python markdown extension enabled? There is also now a tool in the repo that can spit out the versions for you.
  • git hashes now printed in notebook for all relevent software repos
• General Functions: why have you added the kaiser6 window? What is its utility?
  • removed
• Single rlzn: nu_e is surely defined in hera_pspec somewhere, right?
  • yes, in conversions.units. They are the same number, notebook unchanged
• General comment: working needs more comments. Most methods are reasonably easy to understand, but not all of them. I've never used hera_pspec, and I can "generally" follow, but, stuff like keys = uvp_avg.get_all_keys() is a bit opaque. Just desribing with comments/markdown in words what you're doing will help (also realising that pspec API may change over time, and it might get harder to just go and check what each method does).
  • added some more comments, and removed a few stray lines that were not used in the final notebook state
• In[22]: not clear what this is showing.
  • not sure what this was now, its probably deleted now
• Aliased power spectrum: Something weird is going on with the equations, I'm seeing in there.
  • seems like a github rendering problem? it looks fine in jupyter or nbviewer
• Figure 5: I think it's more informative to plot the the residuals with error bars given by mean standard error (i.e. std/sqrt(n)). Also, it would be good to have a horizontal gridline at 0, as well as those at 1%.
  • I think its fine as-is. added line at zero
• It would be good to have text explaining why the 2-sigma line is 7% (it's kind of explained in the Summary).
  • added words in section 7
• I think having a number of sigma within which the results are consistent in the Summary (point 1) will be useful secondary info (not necessarily part of the criteria, just in the Summary).
  • Added a plot in section 6 showing this

-changed additional notes -added scripts used to generate the multiple visibility realizations and associated power spectrum estimates

[zacharymartinot]

Well, for posterity, I have included all of the code and meta data (e.g. random seeds) that went into producing the data products, so that it should be technically possible to rerun everything. But it may require significant effort, and theres nothing to be done about that at this point.