search

pmanion0 / cs224n-pa4

GitHub repository for CS224n's PA4 - Neural Network for Named Entity Recognition
0 stars 0 forks source link

Implement WindowModel Wrapper Around NeuralNetwork #8

Closed pmanion0 closed 8 years ago

pmanion0 commented 8 years ago

Implement all the code necessary to convert from the raw datums to and from the NeuralNetwork class for scoring, training, testing, etc.

Tasks:

pmanion0 commented 8 years ago

I finished the first draft of the training call for the WindowModel. However, right now, it will just run through the dataset once and then terminate. We will obviously need to implement some kind of convergence criteria.

I'm thinking that the WindowModel class should make the decision on when the model is converged, but the NeuralNetwork class should return the information WindowModel needs to make that decision (e.g. error rate on last example).

Thoughts?

pmanion0 commented 8 years ago

Finished both training and testing sections in fefadfc. I'm going to need to create a mini NeuralNetwork model for testing purposes before I can do anymore unless you expect to have a baseline version of it working soon @dxl0632 .

dxl0632 commented 8 years ago

I don't think I can have it soon. My current plan is to finish non-gradient related methods during weekdays. I'll need to spend some amount of time to derive gradient and then I'm going to finish up backprop over the weekend, which means I'll have a working model then.

Also, I'm taking the whole week off next week. So I'll be working on PA4 Monday and Tuesday.

dxl0632 commented 8 years ago

@theforager You were able to run it? When I tried, it gave me this

Caused by: java.lang.InstantiationException: cs224n.deep.WindowModel

pmanion0 commented 8 years ago

@dxl0632 Oops, good catch, there was an issue with the initialization for the class. I fixed it with a hack for now, but I'll revisit it later. 0d8853f

There are still other issues with actually running this like error handling for missing words, etc., which I still need to do. I would test your Neural Network thing with unit tests.

pmanion0 commented 8 years ago

Pretty much all the major functions used in WindowModel have been tested (exception the one issue I added), so I will consider this one done. Any other work we can add in new issues.