It’s a trick question until you add machine learning. As of today the LA Dodgers have a 97-56 Win-Loss record, and the Cleveland Indians have a 96-57 Win-Loss record.
Is one team more Bayesian than the other?
When you add the condition of machine learning, it sheds light.
The concept of recency.
Recency – What Have You Done Lately?
There’s a famous Bayesian analogy with C3PO and Han Solo in Star Wars: The Empire Strikes Back.
As Han Solo, Chewie, C3PO, and Princess Leia are escaping…the Imperial Fighters are hot on their tail. Han Solo decides to escape by flying right into an asteroid field.
The odds of survival according to C3PO: 1 in 3,720.
But those odds are approximate…not specific to the pilot at hand: Han Solo.
Han Solo figures he’s a superior pilot, and the Imperial Fighters will turn around, electing not to fly in the deadly asteroid field.
Solo guessed his own odds of survival are about 75%.
Bayesian modelling, updates with posterior probability…given new information.
Machine learning + Bayesian = processing that new information with an update of every conversion, or every asteroid that you pass.
Bayesian’s wonderful, but if it isn’t machine-learning Bayesian…you’ll likely only use two models a year (and never take advantage of new, recent information in a fast changing environment).
Now enter the Cleveland Indians, managed by former Red Sox manager Terry Francona.
The last part of the summer, the Indians went on a historic 22 game winning streak winning 27 of 28 games, a stretch matched only by the 1884 Providence Grays.
Now, let’s look at the LA Dodgers.
Sports Illustrated put the LA Dodgers on the cover of its magazine August 28 with the headline reading “Best. Team. Ever?”
The Dodgers then led the NL West division with a historic 21 game lead over the next team in their division.
That lead has never happened in the history of the game…ever.
The Dodgers fell from grace….fast. The lead now tallied just nine games.
Both teams have similar Win-Loss records, but looking at recency through the lens of Bayesian posterior probability, the Indians are now the equivalent of Han Solo (actually better).
The Indians have taken off like a rocket ship.
The Dodgers…they would not be the horse to bet on.
The update of new information with Bayesian posterior probability illustrates their very different paths.
Now enter Uber self-driving cars in Pittsburgh.
They utilize Bayesian machine-learning. Posterior probability, but machine-learning added to it.
If the right turn at a bridge results in side-swiping the bumper off a car…that data point (a very recent data point) influences the model in a way that it hasn’t before.
New information + recency plugged in with Bayesian and then cranked with machine learning = best information.
In baseball, the recent information from the Indians and Dodgers shape the model going forward. The same with hurricane paths: Irma, Jose, Maria.
Is it a trick question: Who’s More Bayesian?
Yes. Machine learning is a huge component.
In the end, it’s Uber self-driving cars. More data (millisecond by millisecond), more processing power.
How old are your models?
How young is your data?
Lets’ play ball.