[MADLIB-837] "Value is out of range for type integer" error when calling rf_classify

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Unresolved
• Labels: TriagePending
• Assignee: Hai Qian
• Reporter: Ronert Obst
• Created: 14/Jun/14 18:39
• Updated: 29/Dec/14 18:49

Description

I get the following integer is out of range error if tree depth / number of trees above some unknown threshold when running rf_classify (in this case 500 trees and tree depth of 20):

Error: 22003: value "3300961000" is out of range for type integer
CONTEXT: PL/pgSQL function "__treemodel_classify_internal" line 154 at execute statement
SQL statement "SELECT madlib.__treemodel_classify_internal ( $1 , $2 , $3 )"
PL/pgSQL function "rf_classify" line 45 at assignment
SQL statement "SELECT madlib.rf_classify ( $1 , $2 , $3 , 'f', 0 )"
PL/pgSQL function "rf_classify" line 4 at assignment
LOCATION: pg_atoi, numutils.c:78

With less trees and less tree depth on the same data no error occurs (100 trees and a tree depth of 10):

1. SELECT
   [more] - > *
   [more] - > FROM madlib.rf_classify('rf_results',
   [more] ( > 'rf_test',
   [more] ( > 'rf_test_results'
   [more] ( > );
   -----------------------------------+

   input_set_size

   classification_time

   -----------------------------------+

   6601922

   01:25:39.613089

   -----------------------------------+
   (1 row)

Time: 5142356.074 ms

rf_classify also runs way too slow. 1.5 hours for 6.6 million rows?