Add docstrings

utils.py

@@ -2,10 +2,18 @@ import numpy as np
 
 
 def one_hot_from_indices(index_ids):
+    '''
+    Turn an n-dimensional tensor of integers into an (n+1)-dimensional array of one-hot vectors, containing a 1 in the given index and 0 everywhere else.
+    '''
     return np.eye(index_ids.max() + 1)[index_ids]
 
 
-def make_one_hot(array, axis=-1):
+def soft_to_hard(array, axis=-1):
+    '''
+    Turn an array of floats (perhaps probabilities) into a one-hot vector, with each vector along `axis` having its max value turned into a 1, and the rest turned to 0s.
+
+    Only tested for axis=-1, as this is usually what we want to do.
+    '''
     return one_hot_from_indices(array.argmax(axis=axis))