Todd Lipcon

Consistently feel like stuff is dumbed down to the point where I don't know what's actually happening... Wondering if I should skip Arduino and use FreeRTOS instead for a more familiar environment as a professional c++ programmer

Venturing into Arduino land for a side project. Amazingly easy to get going and build a simple embedded device these days compared to when I last looked at this 15+ years ago! But the ecosystem kind of reminds me of npm where half the library code is very messy

Another application to ranking: some expensive model gives a score to an item and I want to get the top 10. Once I have ten candidates I only want to fully score more candidates if there is a decent chance they will be better than my current worst one

Example application: I have some classifier for spam but it's very expensive. I would like a very cheap variant to run first which quickly passes through the obvious examples, and only run the real model on borderline cases

ML smarties: anyone aware of research like this: given some DNN with a single output (regression/classification), can you train a much smaller pruned variant that only makes errors in a particular direction and with some bound on frequency or magnitude?

Any ideas for the best lock-free structure for a multi-producer single-consumer *set*? (ie a queue but where duplicates are expected and can be coalesced into single insertions across a large time range)

https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/socc10-spikes.pdf … is a good one, but wondering if there are any other recommended ones

Anyone aware of a database workload modeling/benchmarking paper that accounts well for "range suddenly becomes hot but then trails off in popularity" hotspots?

Can now suggest spanner with a straight face for side projects, prefunding startups, etc

One of the coolest things about Spanner is that you can scale more or less indefinitely with strong semantics. But previously the scaling was only in the up direction, and the price of entry was prohibitive for small use cases. No more!https://twitter.com/GoogleCloudTech/status/1397648568303894530 …

My current go to for this is R, with ggplot, but I have to imagine there's something slightly nicer for the slice-and-dice-and-facet-and-color-and-shape that I want to do.

Data science-y question: let's say I have a few thousand rows of data, each row containing a few independent variables and one or more dependent variables. I want to interactively explore to understand trends of how each independent variable effects the dependent one. Best tool?

It bewilders me that my Roomba is smart enough to drive around and mop the floor without falling down the stairs, yet isn't smart enough to know whether or not its tank is full of water.

So, it's 2021, and my Windows VM decided to automatically reboot in the middle of the night, and somehow didn't tell TurboTax to save, and somehow TurboTax has no autosave, and my hours of obnoxious tax return work are gone. flashbacks to high school history papers corrupted