Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
[recommenders-dev] first draft of model coverage reports

Hi all,

you may have noticed that recent versions of recommenders sucked-away (too) much of your memory. This is caused by the increased code base we used for model generation and unfortunately the mining algorithms applied previously didn't scale well with that grow of data. I'm currently applying Sebastian's work on code clustering to shrink the models which I hope will be done until RC2.

However, during this work I felt the need for some reports that allow me to gain some insights into the raw data and the models generated from it. I've created a first draft of two of such reports (number of object usages + number of recommendation patterns per type) for SWT [1] and JRE [2]. I'd like to generate a few more such as 

* raw number of object usage patterns, 
* evaluation performance (e.g., F1) for frequent recommendation scenarios
* model size in terms of 
  * number of doubles per model, and
  * size per model on disk

I'd also like to generate some 'help pages' that show some commons usage patterns per type (s/t like If you use a text widget, you typically use it like this...).

If you have any ideas on this, I'd be glad to take your comments and requests.


Thanks,

Back to the top