Discussion about this post

User's avatar
Daniel Russo's avatar

It would be nice if the greedy method was indeed optimal. If you're running a recommender system, just recommend greedily and you'll automatically discover which content people like! I think this is too good to be true, unfortunately. If you act greedily, you really will insufficiently explore your catalogue of options.

I think the papers that claim otherwise are interesting, but make assumptions that are less innocuous than they appear. For instance, the assumptions in this paper rule out inclusion of an intercept term, or nested categorical variables, in a linear regression.

https://web.stanford.edu/~bayati/papers/greedy.pdf

Expand full comment
Damek Davis's avatar

The link pointing to your book is broken.

Expand full comment
2 more comments...

No posts