jennykim1016 / jjj Goto Github PK

• Propose a model and an algorithm for tackling your task.
• You should describe the model and algorithm in detail and use a concrete example to demonstrate how the model and algorithm work.
• Don't describe methods in general; describe precisely how they apply to your problem (what are variables, factors, states, etc.)?
• You should also have finished implementing a preliminary version of your algorithm (maybe it's not fully optimized yet and it doesn't have all the features you want).
• Report your initial experimental results.

• https://nlp.stanford.edu/projects/glove/ glove word vectors
• softmax : probability of each class, accept all labels that pass certain threshold

• Janel: Naive Bayes Algorithm for the baseline
• Jenny: Linear SVM for the baseline
• James: Come up with an algorithm (Hansohl suggested to use RNN (time-series)) and think of a way to use it (should be able to feed the data in and get the results I think)

• Project proposal [p-proposal](due Thu Oct 26)
• Project progress report [p-progress](due Thu Nov 16)
• Project poster session [p-poster](due Mon Dec 4)
• Project poster session (peer review) [p-peer](due Tue Dec 5)
• Project final report [p-final](due Fri Dec 15)

http://scikit-learn.org/stable/auto_examples/classification/plot_classifier_comparison.html

Overall good proposal, but it really is helpful to have implemented your baseline and oracle (at least in part) - this will help inform you (and us) whether your project scope seems good. I do think it has a lot of space to explore - particularly if you treat the abstracts as time-series ordered word vectors instead of a bag of words (e.g. Naive-Bayes). You could probably get a pretty good baseline with a bag-of-words, which-words-are-present, statistical approach. Moving forward, it might be worth exploring a simple RNN architecture (you can grab word-vector embeddings online to make this easier).