[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [recommenders-dev] Weekly Report

This comes a little bit out of the blue. So I'll provide a little bit more details ;-)

Michael is going to renovate the network (de-)serialization in Recommenders. First changes improved the network initialization speed (w/o IO) by 30%. Michael is now working on another improvements by changing the binary format. These changes target the Recommenders 2.1 Release in February.

Regarding the ominous "sorter" - I'm not sure I posted it here before:
We (Codetrails) build a content assist proposal sorter on top of the Recommenders 2.0 APIs which aim to provide a better proposal sorting than JDT offers right now. They are not officially part of the Recommenders tool suite but I you'd like to give it a spin, an installation guide is available here [1]. Olav is going to work on this in the next weeks.

Olav, Patrick, Stefan, Timur, Madhuranga, Kavith, 
I'd like to encourage you to continue writing the weekly emails and report briefly which new feature is implemented. I'm aware that the progress will be smaller than during GSOC but it will help us to see which projects are still alive and which projects are cooling down.


[1] https://docs.google.com/a/codetrails.com/document/d/1hQRGCbR5IREQfWHzc0q3S68A1lCBX8AFW8I4fNrDXDM/edit

Am 21.10.2013 um 10:22 schrieb Michael Kutschke <kutschke@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>:

What have i done this week?

Implicit feedback for the Proposal Sorter to allow statistics about how good the sorting was compared to the JDT completion. Already merged in the master branch. A change that also records the sorting of the hippie completion and bare Code Recommenders is in Gerrit. Also, there is a new binary serialization format underway for Jayes. I have also started the second approach to make an evaluation component for the CtrlFlow Miner, which will allow the evaluation of our models. As of now, it simulates the evaluation by resolving the test data artifacts and models, deserializing the models, but then just printing the number of usages that would be used for evaluation if it was for real.

What are my plans for this week?

There are still some open suggestions by Andreas wrt. the serialization format. I will investigate those. Also, the evaluation component needs to be filled with actual content.
recommenders-dev mailing list