Github user iyerr3 commented on a diff in the pull request:
https://github.com/apache/madlib/pull/295#discussion_r203811384
--- Diff:
src/ports/postgres/modules/recursive_partitioning/decision_tree.py_in ---
@@ -2327,6 +2328,110 @@ def _tree_error(schema_madlib, source_table,
dependent_varname,
plpy.execute(sql)
# ------------------------------------------------------------
+def _validate_var_importance_input(model_table, summary_table,
output_table):
+ _assert(table_exists(model_table),
+ "Recursive Partitioning: Model table does not exist.")
+ _assert(table_exists(summary_table),
+ "Recursive Partitioning: Model summary table does not exist.")
+ _assert(not table_exists(output_table),
+ "Recursive Partitioning: Output table already exists.")
+
+def _is_model_for_RF(summary_table):
+ # Only an RF model (and not DT) would have num_trees column in summary
+ return columns_exist_in_table(summary_table, ['num_trees'])
+
+def _is_RF_model_with_imp_pre_1_15(group_table, summary_table):
+ """
+ Check if the RF model is from MADlib < 1.15. The group table for
+ >= 1.15 RF models should have a column named
impurity_var_importance
+ if it was learnt with importance param True.
+ """
+ _assert(table_exists(group_table),
+ "Recursive Partitioning: Model group table does not exist.")
+ # this flag has to be set to true for RF to report importance scores.
+ isImportance = plpy.execute("SELECT importance FROM {summary_table}".
--- End diff --
Our convention is to use snake case for Python (i.e. `is_importance`). I
also suggest changing it to `is_importance_set` to make it more explicit.
---
