Started by Ryo Nakamura (@ryo_n) ·
Brier scores, or how to grade a weather forecast
5 sources · 25 min down the hole
How I got here
- Started with
- If the forecast said 70% rain and it didn't rain, was it wrong?
- Ended at
- “brier score vs log loss”
The Trail
- “how do you score a probability forecast” (google.com)
- Brier score (en.wikipedia.org)
invented by a US Weather Bureau statistician in 1950
- Scoring rule (en.wikipedia.org)
- Calibration (statistics) (en.wikipedia.org)
- “brier score vs log loss” (reddit.com)
What I came away with
Not necessarily. You judge probabilities in bulk: of all the days someone said 70%, it should rain on about 70% of them. The Brier score is the mean squared error between the probabilities you gave and what happened (1 or 0), so 0 is perfect and always saying 50% gets you 0.25.
Keep going
- Pakistan's Tekken scene and the 2019 run nobody saw coming · Ryo Nakamura · 5 sources
- the four color theorem was the first big proof nobody could check by hand · Henrik Larsen · 5 sources
- the Delphi method is named after an oracle and came out of RAND · Tomás Rinaldi · 6 sources
- finally understood backprop thanks to 3blue1brown · Arjun Pillai · 5 sources