mjvakili / ccppabc Goto Github PK

Is the only change you made to group_richness.py just the removal of the last bin?

is it possible to simulate an extreme data situation where the results from the standard analysis are not consistent with the ABC ones? if so, how realistic it is?

I get this error when I ran abc_pemcee.py

ValueError: to_rgba: Invalid rgba arg "0.682352941176"
to_rgb: Invalid rgb arg "0.682352941176"
cannot convert argument to rgb sequence

@changhoonhahn can you look at this?

if the Gaussian hypothesis underlying the standard approach are so unrealistic as stated in the text, why results are still consistent with the ABC ones? how likely it is that we will face a real data scenario which might lead to significantly different results?

@mjvakili : should we sprinkle the figures appropriately throughout the text, instead of all at the end?

need a flexible way of building and loading inverse covariance matrices for emcee analyses, this includes changes to data.py

Let's read and understand the relevance of this paper (and important references therein) to our group multiplicity function stuff:

http://arxiv.org/abs/1505.04798

I need to implement this in the code.

Will a real data scenario lead to significantly different results?

if the Gaussian hypothesis underlying the standard approach are so unrealistic as stated in the text, why results are still consistent with the ABC ones? how likely it is that we will face a real data scenario which might lead to significantly different results?

@mjvakili In abc_pemcee.py line 87 you call the ConstEps module from abcpmc. What does this do exactly? Does this keep the threshold eps the same for every iteration or is this a 75th percentile threshold ?

It was originally hardcoded with [1e13, 1e13](commented out on line 86), I'm guessing that this was for testing purposes?

Don't know why! I'll fix it later.

what caveats one would face in trying to apply this kind of analysis in real data? How can they affect the final results and how can they be circumvented?

@mjvakili , in the discussion of Figure 7, 8 in Section 3.4, I'm not sure how meaningful it is to provide a detailed comparison of slight discrepancies between the ABC and Gaussian pseudo-likelihood posterior PDFs.

I think it might be productive to get ongoing discussions on the referee report by raising them as issues. Here's the first item:

a deeper discussion on why the ABC approach should be preferred in this particular context (one or two phrases on this should be included in the abstract). in section 3.4 it is said: Furthermore, the Gaussian-likelihood approach relies on constructing an accurate covariance matrix estimate that captures the sample variance of the data. While we are able to do this accurately within the scope of the HOD framework, for more general LSS parameter inference situations, it is both labor and computationally expensive and dependent on the accuracy of simulated mock catalogs, which are known to be unreliable on small scales (see Heitmann et al. 2008; Chuang et al. 2015 and references therein). Since ABC-PMC utilizes a forward model to account for sample variance, it does not depend on a covariance matrix estimate; hence it does not face these problems.

this might lead to confusion. Do the authors believe that ABC presents no advantage in the specific case studied in this paper?

I need to implement this in the code

@changhoonhahn : just a reminder to remove the 2PCF from this line:

https://github.com/mjvakili/ccppabc/blob/master/ccppabc/text/revision/main.tex#L91

@mjvakili : I think you should specify tab size in your ~.vimrc to be consistent with python syntax. Currently it's inconsistent, so it's pretty confusing.

When I run abcpmc_restart.py on mercer, I get the following ImportError:

Traceback (most recent call last): File "abcpmc_restart.py", line 31, in <module>
import data_multislice as Data
File "/home/chh327/project/ccppabc/halo/code/data_multislice.py", line 9, in <module>
import pyfof
ImportError: dlopen: cannot load any more object with static TLS

I'm not sure what's causing this but I think it might be a good opportunity to do a thorough code review and do a planned submission of all the runs we need for the paper. @mjvakili

@changhoonhahn : could you read what's added to the response to the referee report and edit it?

@mjvakili : I incorporated the sentence you added for the advantages of ABC in accounting for systematics (Section 3.4 Page 10). I added some citations to systematics papers in BOSS. Can you check out the updated paragraph and confirm the changes?

Whenever I do a serious mpirun with many cores, I always get the following message from halotools

...Building lookup tables for the NFWPhaseSpace radial profile.
(This will take about 6 seconds, and only needs to be done once)

No idea what this is about, but we should probably figure out what it is.

@mjvakili : Can you take a look at the paragraph in Section 3.4. that reiterates how Gaussian pseudo-likelihood is wrong? Close this if you think it's fine.

@mjvakili : prettyplot() and prettycolors() are already coded into plotting.py, is there a particular reason you made prettyplotting.py?