def features_importance(self):
query = "SELECT predictor_name AS predictor, ROUND(100 * importance_value / SUM(importance_value) OVER (), 2) AS importance, SIGN(importance_value) AS sign FROM (SELECT RF_PREDICTOR_IMPORTANCE ( USING PARAMETERS model_name = '{}')) x ORDER BY 2 DESC;".format(
self.name)
self.cursor.execute(query)
result = self.cursor.fetchall()
coeff_importances, coeff_sign = {}, {}
for elem in result:
coeff_importances[elem[0]] = elem[1]
coeff_sign[elem[0]] = elem[2]
try:
plot_importance(coeff_importances, coeff_sign, print_legend=False)
except:
pass
importances = {"index": ["importance"]}
for elem in coeff_importances:
importances[elem] = [coeff_importances[elem]]
return (tablesample(values=importances, table_info=False).transpose())
def features_importance(self):
query = "SELECT predictor, ROUND(100 * importance / SUM(importance) OVER(), 2) AS importance, sign FROM "
query += "(SELECT stat.predictor AS predictor, ABS(coefficient * (max - min)) AS importance, SIGN(coefficient) AS sign FROM "
query += "(SELECT LOWER(\"column\") AS predictor, min, max FROM (SELECT SUMMARIZE_NUMCOL({}) OVER() ".format(", ".join(self.X))
query += " FROM {}) x) stat NATURAL JOIN (SELECT GET_MODEL_ATTRIBUTE (USING PARAMETERS model_name = '{}', ".format(self.input_relation, self.name)
query += "attr_name = 'details')) coeff) importance_t ORDER BY 2 DESC;"
self.cursor.execute(query)
result = self.cursor.fetchall()
coeff_importances, coeff_sign = {}, {}
for elem in result:
coeff_importances[elem[0]] = elem[1]
coeff_sign[elem[0]] = elem[2]
try:
plot_importance(coeff_importances, coeff_sign)
except:
pass
importances = {"index": ["importance"]}
for elem in coeff_importances:
importances[elem] = [coeff_importances[elem]]
return (tablesample(values = importances, table_info = False).transpose())