Details
-
New Feature
-
Status: Open
-
Major
-
Resolution: Unresolved
-
6.4.1
-
None
-
None
Description
The current solr-ltr plugin provider two different normalizer implementations, minMax and standard normalizer. The mimMax normalizer doesn't seem to correctly scale the values to fall in between the given min and max. The solr scale function [1] appropriately scales the values to fall in between the given range. But it cannot be used as it is, because it has performance problems and makes the scoring process really slow.
For example if I have a data set [11,51,16,42,18,21] and if we want to scale the values to 1 to 10, I would except the max value in the data set (which is 51) to be normalized to 10, and min value which is 11, to be normalized to 1.
Here is sample output of the minMax normalizer vs Scaling normalizer
| Input | MinMax Normalizer | Scaling Normalizer |
|---|---|---|
| 11.0 | 1.1111112 | 1.0 |
| 51.0 | 5.5555553 | 10.0 |
| 16.0 | 1.6666666 | 2.125 |
| 42.0 | 4.5555553 | 7.975 |
| 18.0 | 1.8888888 | 2.5749998 |
| 21.0 | 2.2222223 | 3.25 |
[1]https://wiki.apache.org/solr/FunctionQuery#scale