Exercise 2.11, expectation-maximisation results

[mloning]

Results from exercise 2.11 seem wrong, as the performance on the test set is very low:

print(eval_viterbi_test)
>>> 0.027530204000792237

I tested code on the current master branch.

I'm not sure what the expected performance is. Discussed the issue with Anton Belyy (@AVBelyy) who brought it to the attention of some of the monitors, just wanted to find out what the error is.