Some of the biggest firms in chicago are completely machine learning driven. Every once in awhile you can find a guy who can do it himself, but usually they have 5-10 phds in a team.
Trust me, there's programming and there's "programming", self-taught programmers are generally the worst (sloppy, inefficient spaghetti code etc.). And then as @wintergasp pointed out, you're probably also looking at needing a decent chunk of algebra/calculus on top of that. You can find all sorts of things on the internet, look on YouTube and you can see people showing you step-by-step serious medical procedures. You could watch as many YouTube videos as you liked on triple-heart bypass operations, but you still wouldn't be any good at them because you would be missing the other 99% of the medical training. Its the same with machine learning and other complex algo trading, you might pick up a bit of programming here and a bit of maths there, but without the rest of the background 99% of people will fail. As @minmike pointed out, the people who "go it alone" and are successful are generally so because either they've got academic background in the subject, or because they are academically gifted in mathematics etc. Someone coming on here saying "Nowadays I want to use machine learning on trading" and "shall I get some tips about this？" does not exactly strike me as fitting into one of the categories listed above. The original post just shouted lack of experience and that's why I gave the reply I did.
thanks man; In the fact I have completed Machine learning course on Coursera teached by Mr Andrew Ng; and I can program by Matlab and Python too. Thanks for your kindness.
Thanks for your reply; Do you mean "Market internals" like what below picture shows(red lines: 15ups and 15downs; black line:6downs-up-6downs): If it is, I have done this but I have a small question which confusing me that if I now open a long order and when / where should I close the order according to this internals or how shall I confirm it is still in this internals structure?
Oh ok cool. Learn c sharp, python is good for academics and basics but you still need a real programming language as its hard to solve complex issues fast with a single threaded language like python. Machine learning is just a component for system selections, focus on algo design first and apply clustering to your systems, then automate the process to constantly have a different system on a weekly or daily basis and ... tada you can use the fancy words machine learning.
I have a the basic foundation of C language and I use Matlab to do ML, so I think that is ok for using ML on trading rather than learning C# which would need much time; I want to use ML to select K bars after which there is a big swing; Do you have some related experience?