This is a must read: http://www.priceactionlab.com/Blog/2015/05/myths-data-mining-bias/ Especially #5
Very interesting. This would actually be an excellent addition to the "Machine Learning is the new C++" thread.
The same author strikes again today: http://www.priceactionlab.com/Blog/2015/05/multi-armed-bandit-problem/ Interesting references to multi-armed bandit problem I wasn't aware of but I guess everyone comes across it in a casino.
I really wish you would feed this into my existing thread. But, you are already doing a good job adding value here. And, its not exactly about machine learning but an alternative. The multi-armed bandit problem is a good example why the retail trader should explore multi-asset trading but within reasonable bounds. It's may be the reason for the growth in options premium selling. Additionally, he highlighted the enduring sources of return which were highlighted Anti Illmanen's book, Expected Returns. The book is very dry. I wanted to share the cliff notes pdf from the net but it disappeared.