DEV: Created standalone ImportanceAnalysis class in

separate wrapper file (#237)

REFACTOR: Updated important_variables and variable_importance
methods to convert to pandas DataFrames

python/varspark/importanceanalysis.py

@@ -0,0 +1,39 @@
+import pandas as pd
+from typedecorator import params, Nullable
+
+class ImportanceAnalysis(object):
+    def __init__(self, _jia, sql):
+        self._jia = _jia
+        self.sql = sql
+
+    @params(self=object, limit=Nullable(int), normalized=Nullable(bool))
+    def important_variables(self, limit=10, normalized=False):
+        """ Gets the top limit important variables
+
+        :param (int) limit: Indicates how many of the most important variables to return
+        :param (bool) normalized: Indicates whether to return normalized importances
+
+        :return topimportances (pd.DataFrame): Dataframe of most important variables containing a
+            variant_id and its corresponding importance.
+        """
+        jimpvarmap = self._jia.importantVariablesJavaMap(limit, normalized)
+        jimpvarmapsorted = sorted(jimpvarmap.items(), key=lambda x: x[1], reverse=True)
+        topimportances = pd.DataFrame(jimpvarmapsorted, columns=['variable', 'importance'])
+        return topimportances
+
+    @params(self=object, precision=Nullable(int), normalized=Nullable(bool))
+    def variable_importance(self, precision=None, normalized=False):
+        """ Returns a DataFrame with the gini importance of variables.
+
+        :param (int) precision: Maximum floating point precision to return
+        :param (bool) normalized: Indicates whether to return normalized importances
+
+        :return importances (pd.DataFrame): DataFrame of variable importances containing variant_id, importance, and split count
+        """
+        jdf = self._jia.variableImportance(normalized)
+        jdf.count()
+        jdf.createOrReplaceTempView("df")
+        importances = self.sql.table("df").toPandas()
+        if precision is not None:
+            importances['importance'] = importances['importance'].apply(lambda x: round(x, precision))
+        return importances