3 ms·
The site appears to be dead, so haven't had the opportunity to actually read it. But I recently picked up a "Machine Learning" project for a client. I don't en
by suvelx 10y ago
The site appears to be dead, so haven't had the opportunity to actually read it.
But I recently picked up a "Machine Learning" project for a client. I don't entirely understand everything in the fullest detail, but I understand the basics. Yet despite my lack of knowledge I got something to work 'enough' in a day. Tensorflow makes getting a working DNN easy, and then wrappers like TFLearn make it nearly trivial.
The hardest part of the project was sourcing a decent amount of training data.
And I'm honestly somewhat ashamed of it. My friends think it's cool, but really I don't feel like I did anything. I just glued a library together.
But maybe that's the interesting bit, that frameworks have been developed to a point that even a rube like me can get something working in a day.
- dasboth 10y agoI think it's definitely a good thing to abstract away the nuts and bolts of a complicated ML algorithm for general use. Implementing the algorithms from scratch is really useful if you want to learn the details, but most of the time the "standard" implementation will do fine, and will probably perform more efficiently. Of course as with any tool, a conceptual understanding is essential to meaningfully interpret the results though. I completely get what you're saying about not feeling like you did anything though. Once you understand the algorithms even at a basic level they stop feeling like magic, even though they'll still feel like that for other people.
- bbctol 10y agoReminds me of that article going around HN a few days ago about the ease of AI programming (http://bjenik.com/AIBasedProgramming/ http://bjenik.com/AIBasedProgramming/) It had a lot of problems, but maybe we really are moving to an era where data analysis consists more of throwing a neural network at it.
- jeffmcmahan 10y agoIt's back now!