Fixed error introduced by 6f58dbf (#2419)

* Fixed error introduced by 6f58dbf in which the scaling variable is initialized as an array instead of a dict. This commit maintains the intentions of the 6f58dbf commit by initializing 'scaling' to a dict of arrays of identity with type floatX.

* Array of ones with shape of selected vars

This commit is actually preserves the intention of the previous commit as it avoids the guess_scaling function. It still maintains that the shape of scaling must be the size of the specified variables instead of the whole model.

* Removed unnecessary function imports to namespace that were previously used for dict to array bijection

Author: ctm22396

pymc3/step_methods/hmc/base_hmc.py

@@ -42,7 +42,9 @@ def __init__(self, vars=None, scaling=None, step_scale=0.25, is_cov=False,
         vars = inputvars(vars)
 
         if scaling is None and potential is None:
-            scaling = floatX(np.ones(model.dict_to_array(model.test_point).size))
+            varnames = [var.name for var in vars]
+            size = sum(v.size for k, v in model.test_point.items() if k in varnames)
+            scaling = floatX(np.ones(size))
 
         if isinstance(scaling, dict):
             scaling = guess_scaling(Point(scaling, model=model), model=model, vars=vars)