[
https://issues.apache.org/jira/browse/LUCENE-7289?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Adrien Grand updated LUCENE-7289:
---------------------------------
Attachment: LUCENE-7289.patch
Here is a work-in-progress, but I'd be happy to have some feedback about how
this is exposed. One known issue is that float to half-float conversion
currently truncates (ie. rounds near 0), but I think it would be better to
truncate to the nearest value like casting doubles to floats does? If someone
has tips how to do it properly, I could certainly make use of some help.
> Half floats
> -----------
>
> Key: LUCENE-7289
> URL: https://issues.apache.org/jira/browse/LUCENE-7289
> Project: Lucene - Core
> Issue Type: New Feature
> Reporter: Adrien Grand
> Assignee: Adrien Grand
> Priority: Minor
> Attachments: LUCENE-7289.patch
>
>
> When it comes to indexing floating-point data, doubles and floats are quite
> space-intensive even though full precision is rarely needed. So maybe we
> should consider exposing half floats?
> Half floats have 5 bits for the exponent and 11 bits (including the implicit
> bit) for the mantissa so the minimum value is ~6x10^−8^, the maximum is
> 65504 and they can represent all integers between -2048 and 2048 accurately.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
