co-occurrence matrix with fixed window of 4 (the default window size), over the Reuters "crude" (oil) corpus.

[Joseph Pareti]

I have a different implementation than yours , and the diagram at 'question 1.5' differs from yours. Do you think the answer should be unique? Perhaps the fact that the co-occurrence calculation for the toy dataset (all that glitters, etc) is correct does not imply that the real dataset is correctly processed?