Closed pmanion0 closed 8 years ago
Implemented in 5ea0b09
Just implemented this as well and will be working on baseline. Do you have any plan to split the work? I don't see a clear way so far.
OK, why don't you implement the baseline, and I'll try and write some interface functions for the neural network. I think once we have some idea of those, we can split up the work.
I think broadly there is the task of implementing all the internals of the neural network (e.g. scoring, back-prop, etc.), another task of implementing all the wrappers around those functions (e.g. run SGD until convergence, etc.), and another task of running all the tests that they asked us to perform (e.g. window size, learning rate, etc.).
If we have clear functions that separate the responsibility of these things, one of us can start working on the later stages even before the gradients are fully working.
The function to read in word vectors needs to be implemented in FunctionFactory.readWordVector and is used in NER.main.
HW Description: You will have to load these vectors into your Java program and save them into a matrix along with their indices and the words they represent. Later, you will compare these word vectors with randomly initialized word vectors.