Michael Clark

📖 Excited to announce my book, Models Demystified, is out Friday! It covers concepts from stats to deep learning, & other topics like uncertainty, causal inference & more! tinyurl.com/book-at-crc (@CRCPress) tinyurl.com/book-web (web) #DataScience #MachineLearning #AI

From classic techniques to cutting-edge machine learning, data science models help uncover patterns and power smarter predictions. Check out our latest blog post for key insights from Michael Clark's book "Models Demystified": bit.ly/4hIwnic

@IsabellaGhement @CRCPress great, thanks @IsabellaGhement !

@statsdatasci @CRCPress Cool! Thanks, Michael! I just made a quick plot showing visually the connection between the three modelling scales for your first movie ratings model using marginaleffects. The code is in the alt description of the attached image.

Hey folks! Been a while since I posted about it, but our book on practical data science modeling has come a long way since then. Would love to hear some thoughts on GitHub or here. Hopefully we'll get it done soon and out on @CRCPress in the near future! m-clark.github.io/book-of-models/

@IsabellaGhement @CRCPress Hi @IsabellaGhement ! A quick note to say thanks again for your thoughts. Like last time I made Github issue with these suggestions so that we'll be sure to keep in mind. I've already incorporated a lot of it too! Thanks!

@statsdatasci @CRCPress Why keep saying “it’s difficult to interpret this effect”? Just show the connection between scales visually.

@IsabellaGhement @CRCPress Thanks for your thoughts @IsabellaGhement . You are right the language is too loose/vague in several spots here (and our footnote is not enough). I've already gone ahead and made several of these and related changes. I'll also revisit a couple other spots also. Thanks again!

@statsdatasci @CRCPress Loved the description of the link function as a bridge between the data and various aspects of the conditional distribution of Y given X. 💪

Link to the repo: github.com/m-clark/book-o…

@JKubale @stephenjwild @StaffanBetner @camjpatrick I posted on fractional and several related options with comparisons here: m-clark.github.io/posts/2019-08-… The fractional logit result can be done as a binomial glm with robust SE (among other ways). I like beta reg if applicable, but that's just me.

@IsabellaGhement We actually had a whole section devoted to odds vs. log odds vs. probability interpretation that was taken out to shorten things up, but maybe that was hasty!

6. It would be helpful to the reader to connect the following: log odds of 0 with the probability of 0.5; log odds > 0 with probability > 0.5; log odds < 0 with probability < 0.5. This can help build more intuition for the log odds beast.

@statsdatasci: As promised, some feedback below on the Generalized Linear Models section of your “book of models”. I started there for now. In no particular order…

@IsabellaGhement @IsabellaGhement Thanks much for your feedback, very helpful! We struggled with how best to present notation quite a bit, so will revisit that. Also the language gets a little too loose at times. @ChelseaParlett I'll make a note of the symmetry too, thanks!

1. Can it be made clearer that GLMs model the conditional distribution of Y given X (as opposed to the marginal distribution of Y)?

@atomasevic Borrow away, and yes those will be stacked.

@statsdatasci Looks great, love the tabbed R/Python code examples. I guess they will be stacked one below the other in the print version? Your Quarto theme is great, I will probably borrow more than a few elements :)

I've been putting together a book on modeling that I hope will appeal to a wide range of audiences, with examples in Python/R. You can check out the in-progress work at: m-clark.github.io/book-of-models/. Hope you find it useful, and feedback is appreciated as we continue to work on it!

@arthurturrell Looks great! Will definitely add to next update!

@statsdatasci This looks great. Cheeky and shameless plug for another intro to Python book for your intro... Python for Data Science (follows the structure and content of the R book): aeturrell.github.io/python4DS/welc…

@RandVegan I think probably because that's what we use more often, and that when we first started out we actually were going to do things mostly 'by hand' as in my doc m-clark.github.io/models-by-exam…. We also were interested in keeping things as similar as possible between the R and Python code.

@statsdatasci looks very clean and easy to digest. I look forward to reading more out of curiosity I saw you didn't use tidy models but rather the direct packages. Given the format of your the book where you compare models / results it would be benefit you - any reason why you didn't adopt it?

@crcgrubbsd Yes, and very much looking forward to it!

I hope @statsdatasci doesn't mind me announcing that this book will be published by Chapman and Hall/CRC likely in 2024.

@IsabellaGhement @crcgrubbsd @IsabellaGhement here is fine but as a GitHub issue or even just email works too!

@statsdatasci @crcgrubbsd I love your prior work, Michael! I have referred to your webpages so often - they are such a valuable resource for applied statistical work. Can’t wait to check out your book. For feedback, where do you prefer to receive it? Here on Twitter?

There are good tools in #rstats (e.g. Robyn) and #python (lightweightmmm), but as noted in the article, you often will just have to roll your own (e.g. via @mcmc_stan or #numpyro).

Been a while, but here's my new post on marketing/media mix models at the @stronganalytics blog. Those used to time series and mixed modeling might find this an interesting application. strong.io/blog/marketing…

Here are some helpful #guidelines for working with tabular #data buff.ly/3ZjT3fI #deeplearning #tabulardata