Update hmm_bach_chorales.ipynb

[karalleyna]

Loss Curves

You can plot the one of HMMs whose num of hidden states is in {4, 8, 16, 24} as follows :

K=16

Model Selection

[murphyk]

• I see that that the NLL on the test set continues to go down as K increases. But surely it will eventually go up if K is too large due to overfitting? It would be good to verify this by adding some larger Ks, eg K=32, K=48