[
https://issues.apache.org/jira/browse/SYSTEMML-554?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Matthias Boehm updated SYSTEMML-554:
------------------------------------
Description:
SystemML natively supports two data types: MATRIX and SCALAR, where matrices
support value type double and scalars support values types boolean, string,
integer (long), and double. With the introduction of 'transform', we already
added a new data type FRAME to the compiler but without full language and
runtime support having transform directly read csv inputs and transform
specification files.
In order to simplify the use of tables with schema on input and output side as
well as in combination with matrices (e.g., SYSTEMML-537 and SYSTEMML-538),
this epic aims to fully integrate FRAMEs into SystemML's language, compiler and
runtime. Beside flexibility on script level, this will also allow for a
seamless integration with other APIs like JMLC and MLContext (e.g.,
SYSTEMML-452), where we do not necessarily consume frames from files but
directly from in-memory objects.
Initially, we will allow the following operations: (1) read/write with a new
read parameter 'schema' (* for all strings, or value type per column), (2)
transform and decode, (3) left/right indexing and cbind/rbind, as well as (4)
casting between matrices and frames. Down the road, frames will also constitute
a key building block for unifying relational and linear algebra (e.g.,
SYSTEMML-439)
was:
SystemML natively supports two data types: MATRIX and SCALAR, where matrices
support value type double and scalars support values types boolean, string,
integer (long), and double. With the introduction of 'transform', we already
added a new data type FRAME to the compiler but without full language and
runtime support having transform directly read csv inputs and transform
specification files.
In order to simplify the use of tables with schema on input and output side as
well as in combination with matrices, this epic aims to fully integrate FRAMEs
into SystemML's language, compiler and runtime. Beside flexibility on script
level, this will also allow for a seamless integration with other APIs like
JMLC and MLContext, where we do not necessarily consume frames from files but
directly from in-memory objects.
Initially, we will allow the following operations: (1) read/write with a new
read parameter 'schema' (* for all strings, or value type per column), (2)
transform and decode, (3) left/right indexing and cbind/rbind, as well as (4)
casting between matrices and frames. Down the road, frames will also constitute
a key building block for unifying relational and linear algebra (e.g.,
SYSTEMML-439)
> Frame data type support
> ------------------------
>
> Key: SYSTEMML-554
> URL: https://issues.apache.org/jira/browse/SYSTEMML-554
> Project: SystemML
> Issue Type: Epic
> Components: APIs, Compiler, Parser, Runtime
> Reporter: Matthias Boehm
> Assignee: Matthias Boehm
>
> SystemML natively supports two data types: MATRIX and SCALAR, where matrices
> support value type double and scalars support values types boolean, string,
> integer (long), and double. With the introduction of 'transform', we already
> added a new data type FRAME to the compiler but without full language and
> runtime support having transform directly read csv inputs and transform
> specification files.
> In order to simplify the use of tables with schema on input and output side
> as well as in combination with matrices (e.g., SYSTEMML-537 and
> SYSTEMML-538), this epic aims to fully integrate FRAMEs into SystemML's
> language, compiler and runtime. Beside flexibility on script level, this will
> also allow for a seamless integration with other APIs like JMLC and MLContext
> (e.g., SYSTEMML-452), where we do not necessarily consume frames from files
> but directly from in-memory objects.
> Initially, we will allow the following operations: (1) read/write with a new
> read parameter 'schema' (* for all strings, or value type per column), (2)
> transform and decode, (3) left/right indexing and cbind/rbind, as well as (4)
> casting between matrices and frames. Down the road, frames will also
> constitute a key building block for unifying relational and linear algebra
> (e.g., SYSTEMML-439)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
