Fix bug in ast parsing for python 2.7

TODO.md

@@ -0,0 +1,41 @@
+- use descriptors for examination
+- things almost end up looking like ASM files. And that's okay!
+- use orderd dicts. they make debugging real damn easy
+- END sentinels (i.e. okay, good, stop processing)
+- Add note on clojure influence
+- Add note: This is mostly experimental. I use this to continue to play in a 
+    open frontiner. But plenty of solutions do this.
+- Pretty HTML failure objects (more than just a dict)
+- naming convention
+- Renaming Effort
+- Failure sets
+- Non-linearized flow (lasso)
+- IPython displays
+- Widget interactors
+- %run -i convention (actually, some magic)
+- Trim execution stack so vaquero is never in path.
+- generate_unit_test
+- smart sampler (named exceptions)
+- tab complete
+- fork-join merge
+    - .n_failures
+    - .merge
+    - does this affect sub-failures?
+- no with pattern
+- refactor unit tests
+- enforce 'key': "value" pattern for my strings
+- if 'k' exists then -> onto value pattern
+- counting collector
+- Assertions in the module (pipeline) make sense
+    - this is about debugging data, debugging by another name
+- use ipython's superreload()
+- use unittest assertions in code, and it works?
+- look at things like python's block comment -- i.e. reuse of """"""
+- for one-off things, the code is not production code. it's held to  a diff 
+  standard. but, that doesn't mean there aren't good proactices, they are just
+  tailored
+- underscores are your friend (conventions. easy to strip. python)
+- THIS PATTERN:
+    y_col = 'accept'
+    X_cols = [k for k in df.columns if k != y_col and not k.startswith('_')]
+- envorimental keys? 'staging', 'dev', 'production'