[
https://issues.apache.org/jira/browse/SYSTEMML-537?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Prithviraj Sen updated SYSTEMML-537:
------------------------------------
Description:
Previous instantiations of text analytics usecases with SystemML used to
communicate feature vectors to SystemML using a 3-column table/relation/matrix
(see SYSTEMML-452 for a detailed description). The first column was defined to
be an integer representing a global feature vector ID. Creating global integer
IDs is hard and time consuming. Can this be relaxed to a globally unique string
ID instead?
SystemML's transform operation will be able to handle the above usecase, but
for reasons related to uniformly handling both features and labels (uniformly
transforming both X and y matrices) we will need to update transform (or
introduce some equivalent functionality) that allows one to apply an already
created recode map and recode from scratch on the same matrix.
was:
Previous instantiations of text analytics usecases with SystemML used to
communicate feature vectors to SystemML using a 3-column table/relation/matrix
(see SystemML-452 for a detailed description). The first column was defined to
be an integer representing a global feature vector ID. Creating global integer
IDs is hard and time consuming. Can this be relaxed to a globally unique string
ID instead?
SystemML's transform operation will be able to handle the above usecase, but
for reasons related to uniformly handling both features and labels (uniformly
transforming both X and y matrices) we will need to update transform (or
introduce some equivalent functionality) that allows one to apply an already
created recode map and recode from scratch on the same matrix.
> Support for text analytics usecases
> -----------------------------------
>
> Key: SYSTEMML-537
> URL: https://issues.apache.org/jira/browse/SYSTEMML-537
> Project: SystemML
> Issue Type: Improvement
> Components: APIs
> Reporter: Prithviraj Sen
>
> Previous instantiations of text analytics usecases with SystemML used to
> communicate feature vectors to SystemML using a 3-column
> table/relation/matrix (see SYSTEMML-452 for a detailed description). The
> first column was defined to be an integer representing a global feature
> vector ID. Creating global integer IDs is hard and time consuming. Can this
> be relaxed to a globally unique string ID instead?
> SystemML's transform operation will be able to handle the above usecase, but
> for reasons related to uniformly handling both features and labels (uniformly
> transforming both X and y matrices) we will need to update transform (or
> introduce some equivalent functionality) that allows one to apply an already
> created recode map and recode from scratch on the same matrix.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
