Small fixes for expected context transformer model

[vianxnguyen]

Description

This PR includes small fixes to the Expected Context Transformer Model to account for changes in numpy and sklearn dependencies.

1. Change use of sklearn normalize to ensure we are passing in numpy array and not numpy matrix since use of normalize and other linear algebra operations for matrices are being deprecated here.
2. Set allow_pickle field to True when loading Expected Context saved files since they contain objects. This field is now default to False as of numpy version 1.16.3 here.

How has this been tested?

Manually.

Other information

Note the change in np.load may affect loading of other models if they save pickled data.

[seanzhangkx8]

KC and I tested locally with notebook, worked with older version of scipy. Will file an issue later about newer version scipy related issue. Thank you Vivian! Will merge now.