Twitter | Pretraživanje | |
Peter Baumgartner
Data Scientist by day, sleeping by night.
1.184
Tweetovi
430
Pratim
1.077
Osobe koje vas prate
Tweetovi
Peter Baumgartner 6 h
Odgovor korisniku/ci @pmbaumgartner
Some redefinitions I learned: Experiment: anything new that's different than the status quo Replicable: not replicable Data-driven: feelings-driven
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 6 h
This article was the ultimate test of rage induction for me. I had to quit reading after this quote.
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 22 h
Odgovor korisniku/ci @vboykis
I had been using .env files and `python-dotenv`. Went through the Kedro tutorial today, it takes the gitignore'd YAML approach, but also does a bunch of other handy things if you're using multiple data sources. A+ experience so far.
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 5. velj
For some reason I never see get brought up when there's conversation about tech ethics... 🤔
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 4. velj
Going through the Prodigy docs when all of the sudden I get hit out of nowhere with this cute mf
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 3. velj
Odgovor korisniku/ci @pmbaumgartner
I also created an app to serve as an example of some of these concepts. It uses data from a research paper that evaluated THC and CBD lab measurements of cannabis products in Washington State.
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 3. velj
Odgovor korisniku/ci @pmbaumgartner
The new stuff includes two themes for Altair that match the default theme of streamlit apps.
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 3. velj
I combined the content from my previous "Intermediate Streamlit" article plus some new notes I'd taken into my own "Streamlitopedia". If you're using , maybe there's something helpful in here for you.
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner proslijedio/la je tweet
Hilary Parker 31. sij
Odgovor korisniku/ci @hspter
The way we talk about data science and focus so much on methods, we actually incentivize working with *bad* data, rather than spending the time to collect good data and then use easy methods with it
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 2. velj
Snarky, but some good points. Reminds me of Gelman's time-reversal heuristic.
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner proslijedio/la je tweet
Ivan Bilan 23. sij
- GitHub Repo Spotlight №6 NLP library that incorporates many Deep Learning-based models into one easy to use package called gobbli:
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 31. sij
Odgovor korisniku/ci @tiagotvv @tdhopper
This guy got all the good ones. +1 for Fairness (or Algorithm Bias), Explainability/Interpretability. Others for challenge mode: - GreenAI / "AI is as bad as X cars" - Storytelling
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner proslijedio/la je tweet
Megan Stevenson 23. sij
I have a new paper with , forthcoming in AEA P&P "Algorithmic Social Engineering" We apply classic strategic communication models to "fair machine learning". In a nutshell: nudging people to change behavior by tweaking an algorithm is hard! 1/
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner proslijedio/la je tweet
Richard McElreath 30. sij
The story of this paper is Paul & I wanted to highlight: (1) how opaque inference is to most scientists (is essentially superstition) (2) how bad inferential methods can become normative So the paper combines both. I wrote the title. Paul did the rest.
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner proslijedio/la je tweet
Brian Nosek 22. sij
"Science is messy, and the results of research rarely conform fully to plan or expectation. ‘Clean’ narratives are an artefact of inappropriate pressures and the culture they have generated." Fabulous editorial from . More editors sign on?
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 28. sij
Odgovor korisniku/ci @BecomingDataSci
I'd usually reach for a bar chart, but I've come across situations where the conditional nature of "how is it wrong" is important. I'm also inspired by the literature suggesting that some reasoning is easier with frequencies rather than percents ()
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 28. sij
Odgovor korisniku/ci @IgorBrigadir
I suppose I could find the most densely populated cell, calculate the distance between points, and rescale the others down to that distance.
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 28. sij
Odgovor korisniku/ci @IgorBrigadir
I didn't think too much about the layout of the points. Right now I'm doing circle packing, which outputs coordinates for a scatter plot, but results in the wide spread for low count cells. Any other layout ideas?
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 28. sij
Odgovor korisniku/ci @pmbaumgartner
Same thing with 100 total points and color.
Reply Retweet Označi sa "sviđa mi se"
Peter Baumgartner 28. sij
Anyone know of any work visualizing confusion matrices by density or count? I feel this is a nice way to get a holistic view of classifier performance and understanding how relatively often each outcome happens. (300 points plotted below)
Reply Retweet Označi sa "sviđa mi se"