fix all nans in GrangerAnalyzer by transposing into lower triangle

This is to fix issue #173 (comment).

`GrangerAnalyzer` put its outputs in the upper triangle only, which
would be removed by `drawmatrix_channels()` which takes only the lower
triangle, creating an array of all `nan`s.

This fix transposes outputs of `GrangerAnalyzer` into the lower
triangle. This also implies that the element (i,j) in the `causality_xy`
attribute denotes "causality" from row-i to column-j.

Author: Xunius

nitime/analysis/granger.py

@@ -108,8 +108,10 @@ def __init__(self, input=None, ij=None, order=None, max_order=10,
             # non-same i's and j's:
             x, y = np.meshgrid(np.arange(self._n_process),
                                np.arange(self._n_process))
-            self.ij = list(zip(x[tril_indices_from(x, -1)],
-                          y[tril_indices_from(y, -1)]))
+            # index into the **lower** triangle. Then the element (i,j) stores
+            # the causality from x to y, e.g. gc['gc_xy'][i, j] = f_x2y
+            self.ij = list(zip(y[tril_indices_from(y, -1)],
+                          x[tril_indices_from(x, -1)]))
         else:
             self.ij = ij