Can anyone here recommend a good open source/free machine-learning software package? Currently I do financial analysis for a living, and at some point I'd like to move into more of a stock-picking role. Towards that end, I'd like to start experimenting now with machine learning to build my own algorithms for market timing.
The problem with trying to beat the Street is that the big players already have very sophisticated trading algorithms from machine-learning analyses, and even if I could match their algorithms, I could never match their trading speed, so I'd always be late to move and miss the opportunity, if my algorithms were built on the usual pieces of publicly-available data (stock values, prominent macroeconomic stats, etc.)
However, I have this idea about a bunch of data inputs that the big players probably aren't factoring in, since they're not directly related to the markets that they're trying to predict, but could still factor in very strongly in an indirect way. They're the kinds of things that analysts for the big players would probably be embarrassed to suggest, since they're a bit out of left field, and so they probably haven't been subjected to machine learning to find patterns. That may mean some opportunity for a very small player to capitalize on blind spots of the big players.... to have code that perceives things that are in their blind spots, letting me make moves they can't beat me to. There's precedent for people succeeding along those lines with hand-made algorithms, but in theory machine-learned algorithms should be even stronger.
I understand the basics of machine learning and I'm an auto-didact, so I think I can figure it out, once I have the software package to play around with. So, if someone knows of a good software package for me to use, that would be great. In the beginning, I'll be working with historical data sets, so I don't need something with a lot of connectivity to get started -- I can worry about coding that later, if I manage to come up with something that seems to beat the Street pretty consistently.
Thanks for any help you may be able to provide.
The problem with trying to beat the Street is that the big players already have very sophisticated trading algorithms from machine-learning analyses, and even if I could match their algorithms, I could never match their trading speed, so I'd always be late to move and miss the opportunity, if my algorithms were built on the usual pieces of publicly-available data (stock values, prominent macroeconomic stats, etc.)
However, I have this idea about a bunch of data inputs that the big players probably aren't factoring in, since they're not directly related to the markets that they're trying to predict, but could still factor in very strongly in an indirect way. They're the kinds of things that analysts for the big players would probably be embarrassed to suggest, since they're a bit out of left field, and so they probably haven't been subjected to machine learning to find patterns. That may mean some opportunity for a very small player to capitalize on blind spots of the big players.... to have code that perceives things that are in their blind spots, letting me make moves they can't beat me to. There's precedent for people succeeding along those lines with hand-made algorithms, but in theory machine-learned algorithms should be even stronger.
I understand the basics of machine learning and I'm an auto-didact, so I think I can figure it out, once I have the software package to play around with. So, if someone knows of a good software package for me to use, that would be great. In the beginning, I'll be working with historical data sets, so I don't need something with a lot of connectivity to get started -- I can worry about coding that later, if I manage to come up with something that seems to beat the Street pretty consistently.
Thanks for any help you may be able to provide.
