ignite-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dmitriy Setrakyan <dsetrak...@apache.org>
Subject Re: [ML] New features and improvement of ML module for 2.7 release
Date Wed, 26 Sep 2018 18:26:54 GMT
Great stuff! Would be nice to see a series of blogs outlining Ignite ML
capabilities.This would give us a better momentum.

D.

On Wed, Sep 26, 2018 at 10:31 AM Юрий Бабак <y.chief@gmail.com> wrote:

> Hello Igniters,
>
> I want to make up some overview of all features and major improvement of ML
> module for this release.
>
> So let me start from the one of our main feature for this release:
>
> *TensorFlow integration* <
> https://issues.apache.org/jira/browse/IGNITE-8670>
>
> This integration allows us to use Apache Ignite as a data source for
> TensorFlow. Also, this integration will allow creating and maintaining
> TensorFlow clusters over Apache Ignite and submit TF jobs to those
> clusters. More details in the related umbrella ticket.
>
> Also, for this release we have some new algorithms:
>
> * Random forest  <https://issues.apache.org/jira/browse/IGNITE-8840>
> * Gradient boosted trees <
> https://issues.apache.org/jira/browse/IGNITE-7149>
> * Logistic regression[binary
> <https://issues.apache.org/jira/browse/IGNITE-8403>][multi-class
> <https://issues.apache.org/jira/browse/IGNITE-8511>]
> * ANN <https://issues.apache.org/jira/browse/IGNITE-9261>
>
> New features related with data preprocessing:
>
> * Pipeline <https://issues.apache.org/jira/browse/IGNITE-9158>
> * L1,L2 normalization <https://issues.apache.org/jira/browse/IGNITE-8663>
> * Data filtering for new datasets
> <https://issues.apache.org/jira/browse/IGNITE-8666>
> * Encoding categorical features [OneHotEncoder
> <https://issues.apache.org/jira/browse/IGNITE-8680>][OneOfKEncoder
> <https://issues.apache.org/jira/browse/IGNITE-8664>]
> * Imputer and Binarizer <https://issues.apache.org/jira/browse/IGNITE-8567
> >
> * MaxAbsScaler <https://issues.apache.org/jira/browse/IGNITE-9285>
> * Dataset splitting <https://issues.apache.org/jira/browse/IGNITE-8667>
>
> New features for a model validation:
>
> * Model estimator <https://issues.apache.org/jira/browse/IGNITE-8669>
> * k-fold cross-validation
> <https://issues.apache.org/jira/browse/IGNITE-8668>
> * Param grid for tuning hyper-parameters in cross-validation
> <https://issues.apache.org/jira/browse/IGNITE-8924>
>
> Other features and improvements:
>
> * Model updating <https://issues.apache.org/jira/browse/IGNITE-9387>
> * ML tutorial <https://issues.apache.org/jira/browse/IGNITE-8741>
> * Optional indexing for decision trees
> <https://issues.apache.org/jira/browse/IGNITE-9064>
> * Learning context for trainers(local parallelizing and logging of training
> process) <https://issues.apache.org/jira/browse/IGNITE-8981>
> * Unification of API for feature extractor
> <https://issues.apache.org/jira/browse/IGNITE-8907>
> * Several tickets for removing old unused classes and improvements for code
> coverage and examples [1 <
> https://issues.apache.org/jira/browse/IGNITE-9124>
> ][2 <https://issues.apache.org/jira/browse/IGNITE-9297>][3
> <https://issues.apache.org/jira/browse/IGNITE-9146>][4
> <https://issues.apache.org/jira/browse/IGNITE-9316>][5
> <https://issues.apache.org/jira/browse/IGNITE-9348>][6
> <https://issues.apache.org/jira/browse/IGNITE-8450>]
>
> Sincerely,
> Yuriy Babak
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message