Machine Learning Frameworks For Classification And Recommendation?

[IndigoIdentity]

Machine Learning Services For Classification And Recommendation?

Has anyone had any success with any of ML services / frameworks / APIs which support classification and recommendation for text data either via supervised or unsupervised learning methods?

There are a few use cases that I have for this which need to be accounted for:

-> I have a heap of raw data, the data contain oddities so I need to be able to distinguish between actual data and irrelevant data.
-> I have a heap of random data, it is uncategorised and I want to be able to create categories and have these categories be populated with relevant data.
-> I have a heap of historical data, this is financial in nature and I want to be able to make financial predictions based on daily records within the data.

I have been looking at http://www.datumbox.com/ and https://rapidminer.com/ which seem like they can do these sorts of things but I am unsure if this is actually the case.

So, I wanted to know if anyone has attempted things like this before and what was used to do it?

Thanks in advance

 

[semaphore]

AWS ML is a pretty good platform to use. I have used it in the past for your last point, also for fraud detection.

 

[cguy]

I've done this stuff in the past - we wrote it all from scratch (C/C++), so nothing I can really point to. We used R to prototype primarily though. The algorithms that make sense really depend on what your data looks like and what you're trying to achieve.

 

[IndigoIdentity]

Thanks Semaphore, I didn't think to look at the AWS services.

It looks like it should be able to accommodate for the prediction part for sure. Not too sure about the other use cases but it seems like it should be able to handle this too. I guess I would have to explore the platform to tell if this would work out for me.

As for writing all manners of code to accomplish these tasks. I am there at the moment (well I have nothing that can make prediction) and i'm looking for a smarter option. One where I can just throw it data and it'll know what to do with it without me needing to constantly check if there is anything else that needs to be coded to be accounted for and and.

Well... I'll see you all in some months when I eventually make some sense of this.

 

[nakedpeanut]

This sounds exactly like the pracs we did at Tuks for AI honors.. But we use a custom framework.

 

[[)roi(]]

Google open sourced this; haven't had chance to play with it, but it looks interesting: https://www.tensorflow.org

 

[SlinkyMike]

Big fan of Rapidminer here, had I not sprung for the license then I'd most likely be neck deep in Tensorflow at this point.

I bought RM before Google announced Tensorflow, I'm still keen to get my teeth into it but I can't justify the time investment at this stage.

 

[IndigoIdentity]

You guys are awesome, thanks so much for the feedback.

@nakedpeanut Where did you study and how much did it cost?

@[)roi(] Thank you.

@SlinkyMike This was most helpful. I will have a look at Tensorflow too.