Hi Dave,
Thanks for your initial project proposal. Here's what we got in the proposal -
I know Deep Learning and Neural Nets are not focus of the class but I want to learn about this topic and willing to do some extra research.
To make the project scope something that is doable, given I have no background in DL, one idea is to use a recent http://www.next.ml “Recurrent Neural Networks for General Sequence Leaning” (http://bit.ly/1E6SPeQ) tutorial. This uses the Python Theano library and the movie db (http://stanford.io/1yxYRC2) and blogger corpus (http://bit.ly/1yK3mtF). Time permitting a python scraper could also be used to get more updated text data. Need to do more research on exact specific’s of how modeling can be improved or what can be implemented.
I think either one of the datasets (Movie DB or Blogger) can be good data sources, but I don't think you should restrict your analysis to just doing "deep learning". Can you think about what's it that you want to learn? Without knowing what outcome you want to learn, we will not be able to work towards a learning algorithm for the problem. I would say, stick to the Machine learning methods we will do as part of the program. It will be hard to help for us on Deep Learning, but you can certainly explore it on your own as a side project. But first, can you narrow down to one dataset and one learning problem?
Hi Dave, Thanks for your initial project proposal. Here's what we got in the proposal - I know Deep Learning and Neural Nets are not focus of the class but I want to learn about this topic and willing to do some extra research.
To make the project scope something that is doable, given I have no background in DL, one idea is to use a recent http://www.next.ml “Recurrent Neural Networks for General Sequence Leaning” (http://bit.ly/1E6SPeQ) tutorial. This uses the Python Theano library and the movie db (http://stanford.io/1yxYRC2) and blogger corpus (http://bit.ly/1yK3mtF). Time permitting a python scraper could also be used to get more updated text data. Need to do more research on exact specific’s of how modeling can be improved or what can be implemented.
I think either one of the datasets (Movie DB or Blogger) can be good data sources, but I don't think you should restrict your analysis to just doing "deep learning". Can you think about what's it that you want to learn? Without knowing what outcome you want to learn, we will not be able to work towards a learning algorithm for the problem. I would say, stick to the Machine learning methods we will do as part of the program. It will be hard to help for us on Deep Learning, but you can certainly explore it on your own as a side project. But first, can you narrow down to one dataset and one learning problem?