Longitudinal models #520
Longitudinal models #520
Conversation
Signed-off-by: Nathaniel <NathanielF@users.noreply.github.com>
Signed-off-by: Nathaniel <NathanielF@users.noreply.github.com>
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
|
I think this is close to done now. Tagging @aloctavodia and @drbenvincent since you both expressed an interest in the topic.
I focus on 2 examples of iterative model building and the extraction of the within-individual and across-individual trajectories of growth learned by the model. I've tried to explain the crucial aspects of the model as i understand them without getting too caught up in the confusing vocabulary of fixed and random effects, so i'd be grateful for feedback on this aspect especially. I don't want to add to the confusion around these terms! In the first example i follow the textbook very closely using a normal likelihood and derive parameter estimates akin to those reported in the text. In the second example i go off-reservation and use a gumbel likelihood with censoring which looks more appropriate to me than the Normal likelihood they were constrained to use in the text. I also briefly call out that you can (and probably should) use Bambi for these kinds of models where possible. |
|
Sorry for slow response @NathanielF. Fingers crossed I'll have time to review this weekend, or early next week. |
No problem! Thanks for letting me know. |
|
Giving this one another nudge in case Friday is a good day for this review @drbenvincent? |
|
View / edit / reply to this conversation on ReviewNB drbenvincent commented on 2023-03-11T17:27:55Z First line should have the name of the notebook in the parentheses, not the _title_ of the notebook NathanielF commented on 2023-03-12T13:54:09Z Amended |
|
View / edit / reply to this conversation on ReviewNB drbenvincent commented on 2023-03-11T17:27:56Z Double check the pymc.sampling_jax import line, just include what's necessary. Optionally, if you use PyMC 5.1.0, you can avoid importing these I believe and provide a NathanielF commented on 2023-03-12T13:57:42Z Yeah, i don't need the blackjax sampler. Removed now. |
|
View / edit / reply to this conversation on ReviewNB drbenvincent commented on 2023-03-11T17:27:57Z Capitalize Bayesian
Structure makes it sound like Bambi is used exclusively, but it's only used for one model? NathanielF commented on 2023-03-12T13:59:12Z Adjusted this now to be more explicit that the bambi section is a digression to explain an alternative method. Capitalised Bayesian too |
|
View / edit / reply to this conversation on ReviewNB drbenvincent commented on 2023-03-11T17:27:58Z Ad space after try/except block NathanielF commented on 2023-03-12T13:59:39Z Adjusted. |
|
View / edit / reply to this conversation on ReviewNB drbenvincent commented on 2023-03-11T17:27:59Z Use the NathanielF commented on 2023-03-12T13:59:52Z Hidden |
|
View / edit / reply to this conversation on ReviewNB drbenvincent commented on 2023-03-11T17:28:00Z Use the NathanielF commented on 2023-03-12T14:02:43Z hidden |
|
hidden View entire conversation on ReviewNB |
|
hidden View entire conversation on ReviewNB |
|
adjusted. Note one concern here is the color formatting renders differently to color the whole line rather than just the symbols on the site. View entire conversation on ReviewNB |
|
Hidden View entire conversation on ReviewNB |
|
Adjusted View entire conversation on ReviewNB |
|
hidden View entire conversation on ReviewNB |
|
hidden View entire conversation on ReviewNB |
|
Adjusted View entire conversation on ReviewNB |
|
adjusted View entire conversation on ReviewNB |
There was a problem hiding this comment.
Just a few minor style issues:
- For the multi-line equation block where we have the align environment, if you place the $ symbol before the = or ~ then everything aligns to that. At the moment it's right aligned.
- Add
time seriesto the tags - Consider moving from the Case Studies section to the Time Series section of the repo
- Missing space after full stop for "our model.The"
- Remove commented out line...
# grade = pm.MutableData('grade_data', df_external['GRADE'].values) - Ideally, it would be possible to suppress console output after plotting with
;. But I can't remember if the black formatting removes that. I believe there's a way to suppress black from removing that - Where you do use
sample_numpyro_nuts, can you use thehide-output(cell tag) because we get a whole bunch of progress bar output that we don't need. Or maybe there's a specific kwarg for that sampler to suppress that? - In the rendered website output, there's something really broken about the rendering of the L2 Watermark header, after the References L2 header. I've not double checked, but can probably be fixed by having that in it's own cell. Either way, there's something not working there. More info here https://www.pymc.io/projects/docs/en/latest/contributing/jupyter_style.html#watermark
|
Done. View entire conversation on ReviewNB |
|
Nice. I'll endeavour to do a final pass, reviewing content/clarity in the next few days and we can ship this :) Do feel free to ping me, but it is on my list. |
…eries tag and aligning equations
On this point, i think i prefer it in the case studies section. Time series analysis/modelling is (in my head anyway) a different kind of beast than longitudinal curve analysis. They're clearly, related... but just not the same bucket for me. Can move it if you feel strongly about it, but i prefer it as it is. On the watermark thing i wasn't sure what you meant... it looks ok to me:
|
Perfect. Thanks for your help on this! |
|
Just a small nudge @drbenvincent. |
|
Typo in this sentence "Implementing the model is PyMC is as follows" |
|
I'm wondering if we could get the model math for the "The Uncontrolled Effects of Parental Alcoholism" model. Don't worry about the colour coding if that gets messy. Same for the "Model controlling for Peer Effects" model |
|
After this sentence
could you just spell out how to read the hierarchical term And we don't need the |
|
I think there's one instance of a lower case "bambi", but everything else is capitalised. This is great. Happy to approve after these minor tweaks 👍 |
|
Thanks @drbenvincent! Will get to these tonight. |
|
Made those changes @drbenvincent. Seemed to work in the Review Notebook, but docs build check renders strangely: Probably related to this: #541
|
There was a problem hiding this comment.
So the formatting issue looks like a broader issue, not something specific to this notebook. So happy to approve. Let me know if there are any minor tweaks you wanted to make, otherwise I'll merge.
Thanks again for the contribution, and for bearing with the tardy reviews.
|
Happy with it. Feel free to merge! Thanks again |
Example Notebook on Longitudinal Analysis and Growth Curve Trajectories.
Related to this issue: #508
Example of iterative model construction on longitudinal data following the Willett and Singer book. In the first model we follow the text quite closely, and in the second model we choose an alternative likelihood model and fully difrerent set of priors.