/writing
Writing
Essays on economics, math, programming, and tools. Shorter notes now live in bits. Subscribe to the writing rss.
- Use eCDFs instead of histograms Histograms are familiar, but empirical CDFs are usually a better default for visualizing distributions: no bin widths, easy percentiles, and better behavior in the tails.
- Stop worrying about class imbalance Data scientists worry too much about class imbalance in classifier training sets. If you care about calibrated probabilities, log loss already handles the imbalance correctly.
- Quantifying uncertainty in probability predictions Suppose you're interested in knowing the chances of an event X occuring (e.g., X = " a nuclear strike over any populated area in the year 2019 "). When making predictions a...
- Economic growth via effective regulation I somehow ended up reading this long essay by economist John Cochrane on growth-oriented policy. I really enjoyed it because it takes a hard-headed approach to topics where...
- Double standard: caffeine vs nicotine [...] there is a kind of puritanical view that everything relating to nicotine is bad and harmful and should be stamped on. - Richard West I'm no expert on this topic, but...
- A thought on methodology I've never felt comfortable with the logical positivists' "science is prediction" characterization, primarily because it neglects what I intuitively think of as the heart o...