Ast hash

[cristianoc]

Incorporate ast hash (from the ast without names of identifiers or constants) in the dataset.

Ast hash seen before
  old: 'myString.translate(None, string.punctuation)'
  new: 's.translate(None, string.punctuation)'
Ast hash seen before
  old: 'root.mainloop()'
  new: 'root.mainloop()'

{ "snippet": "foo", "ast_hash": "e5eb2fb5b4ef0783394b2ff2", ... }

guess that data augmentation by (1) partitioning input-output examples by AST hash and then (2) generating synthetic examples by randomly pairing inputs and output with the same hash would lead to a noticeable improvement