Add figure-loss-small-data animint as a test

[siddhesh195]

loss.small.rds and nb.rds stores subset of data from neuroblastoma. Storing the subset helps avoid loading entire neuroblastoma and improves test executing time

Closes #80

[siddhesh195]

Since the PR only adds a test and not any capability, I did not add NEWS item and increase version number. Please let me know if it still recommended

[tdhock]

please use jointseg:: instead of library(jointseg)

[tdhock]

instead of adding another data set to animint2 (which is already almost too big for CRAN), would it be possible to compute these data? if not, a new data set is OK, but please save it as data/something.RData that we can access via data(something,package="animint2")

[siddhesh195]

It is possible to compute these data instead of storing it. My first attempt at computing the data increases the runtime and memory of the test. Currently it takes 1.3 Gb and 600 seconds on my machine. There may be ways to reduce that by optimizing how loss.small and nb are computed. I will spend some time investigating it:
https://github.com/tdhock/changepoint-data-structure/

[tdhock]

would be great to use a simple/small version of the data (instead of the full/large data), as long as it captures the essence of the issue

[tdhock]

not sure how many clickSelects/showSelected values are there, but typically only 2 subsets are necessary for a test

[siddhesh195]

I rewrote the test to compute data instead of storing it. Using a small subset of data reduces the runtime. The sample function controls the size of the subset:
sample(1:.N, 5)

[tdhock]

good start, thanks, please revise

[tdhock]

also it looks like update_axes is commented in one of the plots, is that the one that will stop working if update_axes is included?
Could you please include two plots, one with update_axes, and one without, and test them both?
Also I guess this is not going to be working until after we fix #48 so I guess we can work on that fix in this branch.

[siddhesh195]

also it looks like update_axes is commented in one of the plots, is that the one that will stop working if update_axes is included? Could you please include two plots, one with update_axes, and one without, and test them both? Also I guess this is not going to be working until after we fix #48 so I guess we can work on that fix in this branch.

Yes, the commented update_axes is the one which stops working. The y axis is the one which does not work. I added another plot to the tests such that both with update_axes and without update_axes are tested

[siddhesh195]

The tests successfully runs on my local machine, however fails after pushing to GitHub:
Error in find.package(package, lib.loc, verbose = verbose): there is no package called ‘neuroblastoma’

Let me spend time investigating it further

[tdhock]

you may have to add neuroblastoma to Suggests DESCRIPTION to fix "there is no package called neuroblastoma"
but again it would be preferable if you could create a small simulated data set that replicates the issue (so we could avoid adding the dependency)

[siddhesh195]

you may have to add neuroblastoma to Suggests DESCRIPTION to fix "there is no package called neuroblastoma" but again it would be preferable if you could create a small simulated data set that replicates the issue (so we could avoid adding the dependency)

I rewrote the test to simulate neuroblastoma data set and replicate the issue. The simulated data set uses same fields and a uniform random distribution of values. We no longer have dependency on neuroblastoma package. However two other dependencies to jointseg, penaltyLearning packages were required to help compute the simulated data set

[siddhesh195]

Hi Toby and Yufan,

I was able to remove dependency of neuroblastoma data and also reduce the time required to compute the data instead. Please let me know how to latest commit looks

[tdhock]

Looks like good progress, thanks!
First of all I think the "Selected data and model" plot could be improved by plotting the geom_segment correctly, see below for an example (first segment should start at first data point, last segment should end at last data point etc)

Next I believe we need to add a test that fails.
#80 (comment) says "if I add clickSelects it stops working (but it should still work)." which is a bit hard to understand (sorry) but here is a more detailed explanation.
Adding the geom_tallrect with clickSelects="changes" yields the following viz below, where the limits on the X axis on the last plot are incorrect -- penalty goes all the way up to 5 but it should be smaller (as in viz above)

So please add the following geom to vizWithUpdateAxes$lines

    geom_tallrect(aes(
      xmin=min.lambda, xmax=max.lambda),
      alpha=0.5,
      clickSelects="changes",
      showSelected="pid.chr",
      data=some.selection)

and add a test about the X axis limits on that plot (max X should not always be 5/max over all pid.chr values, it should be smaller when we click on some other pid.chr values).

Also please delete vizWithoutUpdateAxes, since it is irrelevant (the test should be about the update axis functionality).

[tdhock]

Also here is the rendered data viz on the new gallery https://tdhock.github.io/2024-05-22-changepoint-model-selection/ (without the geom_tallrect which causes the problematic X axis)