Isaac Slavitt

Co-founder and Principal Data Scientist at @drivendataorg | advocate for simple, reproducible data science (& occasionally even heed my own advice)

[Isaac Slavitt @isaacslavitt]

"Prioritizing municipal lead mitigation projects as a relaxed knapsack optimization: a method and case study" just published at @ifors_itor: https://t.co/oI6LM3z1jl

Hoping other cities replicate work done in @TheCityofMalden targeting lead mitigation to minimize child exposure.

[Isaac Slavitt @isaacslavitt]

@josh_tobin_ @gantry_ml ... all of which btw DS people "should" care about but virtually nobody is using for their ad hoc/R&D projects (i.e. the bread and butter work for line data scientists 90+% of the time in actual orgs)

[Isaac Slavitt @isaacslavitt]

@josh_tobin_ Shout out to @gantry_ml -- love what you're working on. The big tension in framing the talk was trying to meet DS's where they are w/out trying to simultaneously educate and advocate for MLOps products/practices at the same time (plenty of talks doing that at any data conf atm)

[Josh Tobin @josh_tobin_]

10 commandments of reliable data science

https://t.co/ZW4OsSudfs

[Isaac Slavitt @isaacslavitt]

(By the way, wish I had known about that paper before the talk -- we had a ton of overlap and would have loved to point people to it. Consider this a ringing endorsement!) https://t.co/cUdLVGdOWy
5/5

[Isaac Slavitt @isaacslavitt]

Not often I get to combine experience from being a shipboard watch officer with the field of data science. But my recent @pydataglobal talk on norms for risk mitigation to prevent project failures was one small opportunity to draw some connections
🧵1/5

[Isaac Slavitt @isaacslavitt]

So much about accident analysis can be applied to technical work; specific errors are unique, but root causes are often well understood.
@hspter drew this connection to the work of the work of @sidneydekkercom beautifully in her paper "Opinionated Analysis Development" (2017)
4/5

[Isaac Slavitt @isaacslavitt]

@AndyGCook @englishpaulm @remarkablepaper For me reMarkable's relationship to iPad is like Kindle Paperwhite's—simple e-ink + lack of web/notifications *is* the killer feature. Built-in focus mode when all other devices maximize distraction. Mainly replicates pen and paper plus smooth writing, nice sync experience, etc

[Isaac Slavitt @isaacslavitt]

My talk on more reliable data science research practices at PyData Global 2022 is up: https://t.co/jvje6bFNsg

[Allen Downey @AllenDowney]

The video of my tutorial on Bayesian Decision Analysis, from PyData Global 2022, is available now.

For links to the video, slides, and Jupyter notebook, start at https://t.co/CZnAaEsg1U

[Isaac Slavitt @isaacslavitt]

@acesounderglass Coming to this late, so apologies if people have already mentioned it a million times, but @DouthatNYC wrote movingly of his experience as a skeptic driven by desperation to enter the parallel universe of alt medicine -- in his case for debilitating Lyme https://t.co/QhlXUb5NnB

[Isaac Slavitt @isaacslavitt]

Cool to see @Govlaunch has added @TheCityofMalden's lead service line replacement prioritization project: https://t.co/xljVXDZ1rw

[Isaac Slavitt @isaacslavitt]

@iodave @IDEOVC Congrats, Dave! Excited to see what you build.

[Isaac Slavitt @isaacslavitt]

@remilouf Dealing with indexing when number of data points varies per group. Not bad in PyMC because of tensor indexing, but it's been painful in Stan without ragged arrays.

[Blackjax @blackjax_mcmc]

MEADS is a ensemble chain adaptation for Generalized HMC that uses multi-core CPUs and GPUs efficiently to sample with many chains. 👇

https://t.co/l01jrLRIfF

[BlueConduit @BlueConduitAI]

There are between 6-12 million #leadpipes in use throughout the U.S., built on a legacy of hazardous #infrastructure. Explore BlueConduit's interactive #map to learn more about lead service line replacement: https://t.co/qIQaJJ492L