Hacker News Reader

Comment by _coveredInBees | Hacker News Reader

This is such buzzword powered horseshit. Absolutely nothing about this problem domain requires Deep Learning. Simple regression models if implemented properly can do just fine. Not to mention that sites like Redfin/Zillow have access to far more data and already have pretty good models for the most part. Just because you can throw a Neural net at a given problem, doesn't make it the right thing to do...

I'd be a lot more forgiving if the site that was linked was an actual blog post that contained information on how deep learning helped tackle this problem. But all the site is, is a way to aggregate emails while throwing buzzwords in your face.

View on HN

As much as I like the phrase "buzzword-powered horseshit," there are always opportunities to apply new technologies in a creative way. In this case, it seems obvious to me that NNs will do a good job in identifying latent features that would otherwise slip past a linear model.

Sure, NNs "could" do that...but they are far more likely to just memorize a bunch of stuff when someone who isn't well versed with ML just applies an out of the box solution to a problem, looks at a simplified cross-validation number (I hope) and assumes that NN is better than some simpler model.

If the OP had actually posted a blog talking through things he did and how the NN helped and why he believes it is actually learning something useful and not just learning the noise, then this would certainly be interesting. But I've seen far too many people take powerful, flexible models with a large number of parameters and throw them at simple problems, with insufficient data and thought.

I'd normally give someone the benefit of the doubt...but taking the sum total of the quality of the linked web page, the complete lack of any meaningful information, as well as op's reply to my comment that added no real useful information regarding the actual benefit of throwing deep learning at this, I'm still inclined to stick to my original feelings regarding this post.

Hey there. I found that regression models do okay when the feature-set is simple. And I can see that deep learning seems like overkill but I found that it can create models that does justice to the complexity of real estate.

Does it make it the wrong thing to do? I've never seen anyone so mad at someone for their technical decisions.

Redin/Zillow et al won't show you cap rates.

Yes. It's completely overblown. To me, that demonstrates that the person behind this is either a) ignorant of the fundamentals or b) just riding the hype wave to attract the naive.

There is of course option c). That would be that taking a contrarian view on the technology will provide a strategic advantage to a business. But I don't see how that would apply here.

It's not just technical decisions... the entire site is a single web-page with zero content except for trying to harvest emails. I mean, if you want people to spam such low quality stuff on HN, then I guess that's your prerogative. But I usually enjoy the much higher quality stuff that people post here on HN and I'd hate for the bar to drop this low.

Reply on news.ycombinator.com