thoughts

The two main questions are if I'm implementing loss and accuracy correctly. 
There might also be an issue for if my annotations are correct
also I might be using a screwy idea of accuracy in general, both theory and practice could be wrong


3 CHEERS FOR JOSH IN HIS FIGHT AGAINST JOSH
IN 
THE QUEST FOR THE BUG

Implementation and Theory problems
Annotations
  Practice is good def good. Windows are super sus but ok for now? Yeah fucking sure
  In the future it would be good to generate more data by changing the position of the datapoints
Loss evaluation
Accuracy evaluation

Setting end values to -1 never gets used
We should read up on madmom to see how they did this. EASY - DO THIS WHEN SPENT

Current error

Problem: How many logits is the right amount of logits. 
Iter: There are two options: 100, which is the amount of hidden states, and 2, which is the amount of output states
My current RNN structure is 1-100-2
Is it okay to have this many more hidden states than input/output states?
There should really be only 2 logits.