This post has been de-listed
It is no longer included in search results and normal feeds (front page, hot posts, subreddit posts, etc). It remains visible only via the author's post history.
As often stated in the statistics world, inference is often derived from statistical models. Of course, it has its own problems to the point where ASA would ban p-value reporting in papers in their journal.
But let's say you're facing an audience, it could be your client or the public. Are bagging/boosting/stacking methods as much of a black box methods as neural network or deep learning methods? Or are they more explicable?
If you have any texts, papers, etc. on these topics, please feel free to reference some!
Subreddit
Post Details
- Posted
- 3 years ago
- Reddit URL
- View post on reddit.com
- External URL
- reddit.com/r/learnmachin...