bandit

The problem is that most bandit algorithms assume the conversion rate is constant - i.e., saturday and tuesday are the same.

It strikes me that the corruption of the Indian government is much more akin to a roving bandit.

Whereas in WoW, if someone hits the bandit first, and you help, they get all of the credit.

If you are lucky, you get a stationary bandit that extracts, but does not want to kill the golden goose.

When you look at algorithms like bandit, you see the trade off of exploration vs exploitation.

One can hunt or defend oneself with a rifle and ammunition, trade the ammunition to others, and even become a bandit.

If the local housing market does well, you can make out like a bandit.

That would be quite a lot of work but no doubt the guy who sells the utility makes out like a bandit.

If sat and tues have different conversion rates, this will horribly break a bandit.

I made it through Bandit and Leviathan not too long ago, and I only have one more challenge in Natas.

The stationary bandit wants to act as a gardener and grow production.

You do not want to use a classical bandit for A/B testing.

I was really hoping for an retrospective about a multi-armed bandit learning algorithm that ran in production but never experimented, effectively yielding a "zero-armed bandit".

At some point, someone's bound to say "The emperor has no clothes!", do something that people actually want, and make out like a bandit.

The next thing after A/B testing is the "multi-armed bandit"