There's so much hype and bullshit around Machine Learning (ML). And if I have to read one more crappy prediction of who survived on the Titantic, I'll go postal.

So, what real-world problems are you using it to address...and how successful has it been? What decisions have you supported using ML? What models did you use (e.g. logistic regression, decision trees, ANN)?

Anyone got any boringly useful examples of ML in production?

And don't say you're using it to predict survival rates for the design of new cruise ships...although, to be fair, that might be quite interesting...

The hardest part of making a logistic regression model is to write "threshold" correctly.

A friend who just got into ML recently.

"Dude, did you know how amazing ML is??"
"I'm training a computer to give out outputs, basic AI dude"
"Dude logistic regression is the shizz"
"You heard about backprop mate?"
"ANN is the next big thing. I'm currently working on one of the biggest AI project now"

So I casually ask him whether he completely his project or not. He proudly showed me a 9 lined code he copy pasted from Google (search for neural network in 9 lines) and said, "Dude I trained my laptop with some advanced AI techniques to give out the perfect XOR outputs"

He rounded off values like 0.99 to 1 and 0.02 to 0 to make it look perfect.

#facepalm

Working a week on LSTM based text classifier, getting 89% accuracy only to then get better result with Logistic Regression which was supposed to serve as baseline, lol. Background: 180+ classes of google product categorization taxonomy, 20 million rows of data items (short texts). Had a similar experience once on sentiment classification, where SVMlight outperformed NN models.