Machine Learning

Anyone have some serious experience with Machine Learning? Im into making scrapers / bots for fun and profit and am gonna have a go at learning machine learning to data mine some ecommerce sites (screw the TOS) to see if I can get insights that will be profitable.

Would be cool if I had someone to chat to about it who's also in this community.

It's been a while since I did any AI development but I've previously used a three layer perceptron to watch for any unusual activity within it's domain, but I think a decision tree implementation might work better for you in this case.

seeplusplus said:

It's been a while since I did any AI development but I've previously used a three layer perceptron to watch for any unusual activity within it's domain, but I think a decision tree implementation might work better for you in this case.

Click to expand...

I think you mean 3 layer Neural Network, a perceptron just takes two inputs. Overall it depends what OP is actually trying to accomplish, a decision tree is good for classification tasks but it sounds like he's may not be interested in that. OP, I'd suggest taking the Udemy course from Andrew Ng which is really good!

I built an expert system in prolog a few years back. its a little different than neural nets.

The application of machine learning on various domains is of specific interest to me. I have used sci-py etc. to build these in python. The problem however is the inability to compile the code. I am exploring various ways of using sci-py and compile the "rules" to employ.

If you find a way to compile python, it would be good to know as I think building bots that learn (unsupervised) from content will leave substantial opportunities for the purpose of deriving profit from the web.

I am not very into machine learning, but I was exploring Google's TensorFlow and it is awesome. A lot of examples, quite good comunity and it works with Python and C++

Take a look at it.

I gave neural nets a shot a while back. Worked alright for simple classification, but I found the sample data size required grew too quickly for it to be viable for more complex problems. Not an expert though

I'm currently working on one of those online MOOCs on Machine Learning right now.. basically using Python and Scikit-Learn.One thing for sure, you'll need a HUGE sample size of ecommerce sites to get any meaningful results.. probably at least a few hundred of not thousands of sites or else you'll run into overfitting.Another project I'm considering is to train a neural network or other classifier to figure out which people you follow on Twitter will more likely follow you back than on average. That way, you don't have to mass follow thousands of people everyday (and risk getting banned) but instead follow a smaller pool of people and stay under the radar while you get similar or better followers. This algorithm may need to be retuned for each niche rather than having a one size fits all algorithm for all niches.

I think you must have to go with that. Its a good option to learn machine language. It helps a lot in professional career.

JustCash said:

Let us know what you find out

Click to expand...

Although this is a year old thread, I'd be interested to know if OP and others did develop this. Interesting thread with lots of technical info.