Notes after first read

[CKrawczyk]

• 1
  • I got a bit lost in the introduction, I don't think I know enough about any of the papers cited to follow it very well (or I am just having are hard time concentrating right now)
  • Last two paragraphs of intro seem a bit disjointed and parts almost sound line an abstract (e.g. I don't think results should go here)
• 2.1
  • "...galaxies selected analyses in..." typo
  • Might need an overview of what "validation subset" means for anyone who has not seen the first paper
  • The sentence about scaling the data should be moved to 2.2
• 2.2
  • "...sigma_gal, which we will assume is the same in all galaxies:"
  • you define sigma_r but don't show how it is used
  • I feel like the model definition is incomplete. Make sure all equations in the output of pymc3 display(model) are in the paper (e.g. Lik ~ Normal(r, sigma_r))
• 3.1
  • "...a hich degree..." typo
  • last paragraph is stops mid sentence
• 3.3.1
  • This is the first use of "hierarchical normal model", I feel like this name should be introduced in section 2.2
• 4
  • 4th paragraph "We we" typo
• 5
  • Add Zooniverse acknowledgement

[tingard]

Thanks Coleman, do you think a short section on HMC and Hierarchical modelling is needed in the introduction?

[CKrawczyk]

I don't think so, but I am also very familiar with the techniques. If you want to include a bit more background I would put that information in the modeling section rather than the introduction as you are talking more about the tools rather than the science.

[CKrawczyk]

For my second paper I saved that kind of overview for the appendix as it can easily disrupt the flow of the paper.

[tingard]

I feel like the model definition is incomplete. Make sure all equations in the output of pymc3 display(model) are in the paper (e.g. Lik ~ Normal(r, sigma_r))

Not sure how to make this legible...

[CKrawczyk]

This would be something like:

c ~ Cauchy(\alpha=0, \beta=10)
\phi_{gal} ~ Uniform(0, 90)
\sigma_{gal} ~ InverseGamma(\alpha=2, \beta=20)
\sigma_{r, i} ~ InverseGamma(\alpha=2, \beta=0.5)
\phi_{arm, i} ~ TruncatedNormal(\mu=\phi_{gal}, \sigma=\sigma_{gal}, lower=0, upper=90)
Likelihood ~ \prod_{i=1}^{N} Normal(\mu=r(\phi_{arm, i}, c), \sigma=\sigma_{r, i})

Where N is the number of galaxies in the sample and r(\phi, c) is given by equation ###.

Note, if any of this is incorrect you can fix it, but this is how you can display a multi layer model. The _{i} indexing is the key to make the structure more clear. You can also write the hyper priors in their own block to help keep them separate from the rest of the model.

[tingard]

I think most of these have been addressed in the most recent commits!

[tingard]

Hopefully everything here has been addressed, I'm closing this issue