truncated regression example #30

drbenvincent · 2021-01-24T18:27:46Z

Ok. So this is a fresh (second) attempt at a pull request. I've added a single truncated regression example which I put into a new branch. Hopefully the pull request into main is the right thing to do?

At the moment the notebook does include quite a lot of warnings generated by theano. I'm not sure how to remove them. If someone can let me know then I can update.

review-notebook-app · 2021-01-24T18:27:49Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

ricardoV94 · 2021-01-24T19:09:22Z

The theano warnings will be solved in the next release, so don't worry about them for now. We can wait / rerun the notebook later to hide them :)

ricardoV94 · 2021-01-24T19:10:57Z

Do you want to include the censored regression example in the same notebook? It would be really nice to have a pointer that clearly describes the two types of models (truncated and censored) and shows them side-by-side.

drbenvincent · 2021-01-24T19:33:03Z

Sure thing. Happy to do that. It might make it less repetitive to have it all done in one notebook.

AlexAndorra · 2021-01-24T20:12:11Z

Good idea, let's do that 👌

drbenvincent · 2021-01-25T11:50:18Z

The notebook now covers both censoring and truncation. I've also added more text explanation. Let me know what you think.

AlexAndorra

This is really great and well explained, thanks @drbenvincent 🤩
I spotted some typos and left some suggestions below -- feel free to ask if anything is unclear.
Note that CI is failing because you didn't run the pre-commit steps locally. You can take a look at the PyMC NB style guide for this 😉

examples/generalized_linear_models/GLM-truncated-censored-regression.ipynb

AlexAndorra · 2021-01-25T18:32:33Z

examples/generalized_linear_models/GLM-truncated-censored-regression.ipynb

@@ -0,0 +1,1356 @@
+{


Same comments as for first model
If I understood correctly, the left_censored and right_censored steps could also just be equalities, right? left_censored = (y == bounds[0]), since y can't, by definition, be below/above the bound

Reply via ReviewNB

I thought it might be safer, numerically. But can use equality if preferred

Ah I see! No it's ok, but maybe just add this precision in a comment?

AlexAndorra · 2021-01-25T18:32:33Z

examples/generalized_linear_models/GLM-truncated-censored-regression.ipynb

@@ -0,0 +1,1356 @@
+{


Maybe you can add some pointers to useful resources, for people who are interested?
And also maybe a high level idea of what you would do to get better estimates here on your truncated regression -- and what you mean by "a number of factors"?

Reply via ReviewNB

I've added text about the factors influencing the degree of bias.

Yet to add some pointers to further reading...

I've now added some info at the end of the notebook on further topics and further reading

drbenvincent · 2021-01-25T18:39:50Z

I spotted some typos and left some suggestions below -- feel free to ask if anything is unclear.

Thanks @AlexAndorra. I'm new to ReviewNB. Do I accept/reject these in ReviewNB? Or is that more for discussion then I go implement the changes locally?

Note that CI is failing because you didn't run the pre-commit steps locally. You can take a look at the PyMC NB style guide for this 😉

Will take a look.

AlexAndorra · 2021-01-25T19:11:57Z

is that more for discussion then I go implement the changes locally?

Yes exactly, accepting changes isn't available in ReviewNB -- that'd be a great feature though!

drbenvincent · 2021-01-26T11:07:45Z

I believe all that's left is to add some 'further reading' type notes at the end. I'll also clarify that you can also do censored models that impute the censored values. Should be able to finish off in the next day or so.

drbenvincent · 2021-01-26T14:02:25Z

I think I've addressed all the points now - this is feasibly ready :) Let me know if there are any other final points.

review-notebook-app · 2021-01-26T15:55:15Z

View / edit / reply to this conversation on ReviewNB

ricardoV94 commented on 2021-01-26T15:55:14Z
----------------------------------------------------------------

Small suggestion: Remove the repeated "Hopefully", I would keep the second

drbenvincent commented on 2021-01-26T16:50:22Z
----------------------------------------------------------------

done

review-notebook-app · 2021-01-26T15:55:15Z

View / edit / reply to this conversation on ReviewNB

ricardoV94 commented on 2021-01-26T15:55:15Z
----------------------------------------------------------------

Suggestion: Link to our notebook on censored data in survival analysis, where there is one example of inputing the censored values: https://docs.pymc.io/notebooks/censored_data.html#Model-1---Imputed-Censored-Model-of-Censored-Data

drbenvincent commented on 2021-01-26T16:50:32Z
----------------------------------------------------------------

done :)

ricardoV94 · 2021-01-26T15:56:35Z

@drbenvincent This notebook is really great. I commented two small suggestions but feel free to ignore them :)

drbenvincent · 2021-01-26T16:50:23Z

done

View entire conversation on ReviewNB

drbenvincent · 2021-01-26T16:50:33Z

done :)

View entire conversation on ReviewNB

drbenvincent · 2021-01-26T16:51:30Z

I've implemented those changes - fingers crossed that's everything. Thanks for the tips and suggestions by the way.

review-notebook-app · 2021-01-26T16:55:11Z

View / edit / reply to this conversation on ReviewNB

ricardoV94 commented on 2021-01-26T16:55:11Z
----------------------------------------------------------------

Typo: preactically -> practically

drbenvincent commented on 2021-01-26T16:58:45Z
----------------------------------------------------------------

my spelling is awful! I need to see if there's a spell checking extension for jupyter lab. But resolved it now

review-notebook-app · 2021-01-26T16:55:12Z

View / edit / reply to this conversation on ReviewNB

ricardoV94 commented on 2021-01-26T16:55:11Z
----------------------------------------------------------------

Typo? There are also (followed by nothing)

drbenvincent commented on 2021-01-26T16:58:10Z
----------------------------------------------------------------

done. Well spotted

typos

drbenvincent · 2021-01-26T16:58:11Z

done. Well spotted

View entire conversation on ReviewNB

drbenvincent · 2021-01-26T16:58:46Z

my spelling is awful! I need to see if there's a spell checking extension for jupyter lab. But resolved it now

View entire conversation on ReviewNB

drbenvincent · 2021-01-26T20:05:56Z

Think this is ready now :)

I have to get some teaching and other duties out the way, but I'm happy to take a look at improving the Censored Data notebook in the near future

AlexAndorra

This is really great, thanks @drbenvincent ! I think we should wait for the new Theano-PyMC (or PyMC patch release 3.11.1) to merge -- that'll get rid of the Theano warning flood

review-notebook-app · 2021-01-28T18:03:51Z

View / edit / reply to this conversation on ReviewNB

MarcoGorelli commented on 2021-01-28T18:03:51Z
----------------------------------------------------------------

This might just be me not understanding, but I find this a bit confusing - the lower bound is a horizontal line, so what does it mean to be "to the left of the lower bound"?

fonnesbeck · 2021-01-28T18:15:14Z

@MarcoGorelli I think it is referring to "left" on the number line, rather than a plot.

drbenvincent · 2021-01-28T18:24:38Z

@MarcoGorelli I think it is referring to "left" on the number line, rather than a plot.

Yep. I've used both the left/right and upper/lower terminology. Happy to elaborate on that in the notebook if you think it needs it. Just let me know.

MarcoGorelli · 2021-01-28T19:02:18Z

It's probably clear enough to most people reading, thanks for explaining

Great job on this really nice notebook!

drbenvincent · 2021-01-28T21:09:16Z

I'll see if I can contribute some more example notebooks in the future :) If I have an idea, should I open an issue to run it past people, or just go for it with another pull request?

twiecki · 2021-01-29T16:12:39Z

@drbenvincent You can just write it up 👍.

ricardoV94 · 2021-02-16T16:34:30Z

Pymc 3.11.1 is out. @drbenvincent Do you want to rerun the NB in the latest version to get rid of those annoying warnings?

drbenvincent · 2021-02-17T13:46:00Z

That should hopefully be it @ricardoV94

ricardoV94

LGTM! Thanks a lot @drbenvincent and looking forward for your next PRs :D

drbenvincent · 2021-02-21T10:43:08Z

Is there anything else I need to do to get it live on the site?

ricardoV94 · 2021-02-21T11:49:09Z

Not really. I don't have permissions to merge on this repo, but someone will soon and then rebuild the docs.

AlexAndorra · 2021-02-21T17:00:26Z

All good now! Thanks for the reminder @drbenvincent and for this great PR -- as @ricardoV94, I'm looking forward to the next ones now 🤩

truncated regression example

56682c3

MarcoGorelli self-requested a review January 24, 2021 18:29

now covers truncation and censoring

014f51d

drbenvincent added 2 commits January 25, 2021 11:52

remove irrelevant text

3c0bf62

edits to text

c8b49aa

ricardoV94 mentioned this pull request Jan 25, 2021

pm.Bound and TruncatedNormal generate wrong gradients pymc-devs/pymc#4417

Closed

AlexAndorra requested changes Jan 25, 2021

View reviewed changes

AlexAndorra reviewed Jan 25, 2021

View reviewed changes

drbenvincent added 5 commits January 25, 2021 21:01

fixing pre-commit checks

dfe5ab5

spelling mistakes

3400183

return_inferencedata + remove excess model context

4077aff

variable renaming

3347ce0

comments re inequalities in censored model

289069c

drbenvincent added 3 commits January 26, 2021 13:14

text on what influences level of underestimation bias

5a0312c

fix typo

26be04a

Further topics and reading + references

9bb2e20

fix typo

a61cf17

final tweaks

ad4375d

Update GLM-truncated-censored-regression.ipynb

5df0f64

typos

final (honest) typo correction and output supression

fe108b6

AlexAndorra approved these changes Jan 27, 2021

View reviewed changes

AlexAndorra added documentation don't merge enhancement New notebook planned to be added to the collection labels Jan 27, 2021

run with PyMC3 v3.11.1 to remove warning messages

f0de48c

ricardoV94 approved these changes Feb 17, 2021

View reviewed changes

AlexAndorra merged commit a7d6e5c into pymc-devs:main Feb 21, 2021

drbenvincent deleted the truncated-regression-example branch February 22, 2021 11:13

truncated regression example #30

truncated regression example #30

Conversation

drbenvincent commented Jan 24, 2021

review-notebook-app bot commented Jan 24, 2021

ricardoV94 commented Jan 24, 2021 • edited Loading

ricardoV94 commented Jan 24, 2021

drbenvincent commented Jan 24, 2021

AlexAndorra commented Jan 24, 2021

drbenvincent commented Jan 25, 2021 • edited Loading

AlexAndorra left a comment

Choose a reason for hiding this comment

AlexAndorra Jan 25, 2021 • edited Loading

Choose a reason for hiding this comment

drbenvincent Jan 25, 2021

Choose a reason for hiding this comment

AlexAndorra Jan 26, 2021

Choose a reason for hiding this comment

drbenvincent Jan 26, 2021

Choose a reason for hiding this comment

AlexAndorra Jan 25, 2021 • edited Loading

Choose a reason for hiding this comment

drbenvincent Jan 26, 2021

Choose a reason for hiding this comment

drbenvincent Jan 26, 2021

Choose a reason for hiding this comment

drbenvincent commented Jan 25, 2021 • edited Loading

AlexAndorra commented Jan 25, 2021

drbenvincent commented Jan 26, 2021

drbenvincent commented Jan 26, 2021

review-notebook-app bot commented Jan 26, 2021 • edited Loading

review-notebook-app bot commented Jan 26, 2021 • edited Loading

ricardoV94 commented Jan 26, 2021

drbenvincent commented Jan 26, 2021

drbenvincent commented Jan 26, 2021

drbenvincent commented Jan 26, 2021

review-notebook-app bot commented Jan 26, 2021 • edited Loading

review-notebook-app bot commented Jan 26, 2021 • edited Loading

drbenvincent commented Jan 26, 2021

drbenvincent commented Jan 26, 2021

drbenvincent commented Jan 26, 2021

AlexAndorra left a comment

Choose a reason for hiding this comment

review-notebook-app bot commented Jan 28, 2021

fonnesbeck commented Jan 28, 2021

drbenvincent commented Jan 28, 2021

MarcoGorelli commented Jan 28, 2021

drbenvincent commented Jan 28, 2021

twiecki commented Jan 29, 2021

ricardoV94 commented Feb 16, 2021

drbenvincent commented Feb 17, 2021

ricardoV94 left a comment

Choose a reason for hiding this comment

drbenvincent commented Feb 21, 2021

ricardoV94 commented Feb 21, 2021

AlexAndorra commented Feb 21, 2021

ricardoV94 commented Jan 24, 2021 •

edited

Loading

drbenvincent commented Jan 25, 2021 •

edited

Loading

AlexAndorra Jan 25, 2021 •

edited

Loading

AlexAndorra Jan 25, 2021 •

edited

Loading

drbenvincent commented Jan 25, 2021 •

edited

Loading

review-notebook-app bot commented Jan 26, 2021 •

edited

Loading

review-notebook-app bot commented Jan 26, 2021 •

edited

Loading

review-notebook-app bot commented Jan 26, 2021 •

edited

Loading

review-notebook-app bot commented Jan 26, 2021 •

edited

Loading