PDA

View Full Version : AI library



iacoposk8
02-25-2015, 08:26 AM
hello everyone! what is the best library or sdk (in any language, maybe C ++) to make predictions? as versatile as possible, that allows me to classify, generate text, face recognition, etc ... say a library that would allow me to manage the artificial intelligence in any project.
thank you.

jwatte
02-25-2015, 10:49 AM
There is no single best algorithm to "generate predictions" because "best" typically means "least error between prediction and outcome," and minimizing that error requires a lot of domain-specific mechanisms.

Further, while I understand how you could use a predictor as a classifier ("predict what this input would be classified as") I don't understand how it could be used to "generate text."

There also doesn't really exist any library that will "manage the artificial intelligence in any project," because the kinds of artificial intelligence that exist are very narrow; focused on one task in a known context. If what you want to do is receive sound input, and output possible spoken text, that is an "AI" task that has some libraries. If what you want to do is to first train on a database of faces, and then receive pictures in, and detect and mark recognized faces within those pictures, there are some libraries for doing that, that work very differently from the speech-input ones. If what you want to do is receive some situational awareness through sensors (sound level, LIDAR, whatever) and turn that into a plan-of-action to achieve some goal (navigate-to-position,) there are some sets of libraries that can be combined to do that. Again, they work very differently from the other kinds of libraries mentioned.

The best thing you can do is first to learn what machine learning and AI really is -- what are the terms, what are the areas, and what are the basic approaches. Then you will be able to evaluate various libraries for various tasks. Note that actually understanding machine learning and AI requires that you also understand some pretty sophisticated math (graduate-level statistics and algebra, ideally) so if you don't have that already, you may first have to learn the math to be able to learn the domain to then be able to understand the libraries.

It's probably not what you wanted to hear, but there really aren't any short-cuts around this one.
Maybe try starting here? http://machinelearningmastery.com/4-steps-to-get-started-in-machine-learning/

iacoposk8
02-26-2015, 03:01 AM
I suspected a little. I completely understand, but I was hoping there was a shortcut :)