ignite-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Yury Babak <y.ch...@gmail.com>
Subject Ignite ML news
Date Tue, 10 Oct 2017 18:58:50 GMT
Hi Igniters!

I want to provide some updates about current state of ml module, so let me
introduce that we already have:

* OLS lin regression(not so efficient in distributed case, but Alexey
Zinoviev already working to fix that, IGNITE-6222).
* K-means (IGNITE-5113).
* Decision tree(almost done by Artem Malykh, IGNITE-5218).
* Suitable for distributed calculations block matrices(IGNITE-5791).
* BLAS support (IGNITE-5777).
* And some minors bugfixes/improvements.

As new features now we worked on the several algorithms:

* FCM (Ilya Nozhkin, IGNITE-5246) - Fuzzy c-means, will be the first fuzzy
algoritm in our ML module.
* Log regression (Vladisav Jelisavcic, IGNITE-5059) - logistic regression
and regularization mechanism.
* SVM (Oleg Ignatenko, IGNTIE-6585) - Support vector machine, another
popular classification algorithm.

Also we start works related with neural networks. And here we working on two
big ares. 

The first area is implementation our own distributed neural networks over
Ignite(IGNITE-6386). We will start with CNN(convolutional neural network).

And the second area is integration with existing libs and using them as
backend wrapped on our API. Here we looks on Encog and Deeplearning4j libs.

Also we want to add API for group model trainings. Using this API we will
able to train and choose best model. We even could choose between different
models, like k-mean and decision tree for example.

Any questions or suggestions are welcome.


Sent from: http://apache-ignite-developers.2346864.n4.nabble.com/

View raw message