regargind maemo.org app-karma

Hi, everyone.
I recently started to work with nemein and among other things working
on application karma task from current sprint.

We talked with Niels Brett a yesterday and decided, that formula from here:
http://wiki.maemo.org/Task:Karma_for_applications wouldn't work. It is too
rough and will give a lot of spikes. Instead, I proposed another one.
Here is a little test-spreadsheet for formulas:
http://spreadsheets.google.com/ccc?key=reZ9v3xyJKfKhHAvs9gixqA

Column F has formula which compares average data from latest week with
average data from last week. Also, it uses coefficients to make impact of
parameters smaller. It is not perfect, still (karma can get negative under
some circumstances, but that would be difficult to achieve). I like, that it
gives a smooth curve to karma and helps to highlight apps which recently
became more popular.

Implementation of this formula will require 2 weeks of daily-stats. Niels
will start to gather those (we still need to discuss how those will be
stored).

If you have any ideas on making formula better, I will be glad to
implement those. As soon as we have enough stats-data we would be able
to tweak formula and see results instantly

--
Alexey Zakhlestin
http://www.milkfarmsoft.com/

p.s. Sorry for typos. Wrote this on the run from iPhone :-/

On Friday, May 29, 2009, Alexey Zakhlestin <indeyets@gmail.com> wrote:
> Hi, everyone.
>  I recently started to work with nemein and among other things working
> on application karma task from current sprint.
>
> We talked with Niels Brett a yesterday and decided, that formula from here:
> http://wiki.maemo.org/Task:Karma_for_applications wouldn't work. It is too
> rough and will give a lot of spikes. Instead, I proposed another one.
> Here is a little test-spreadsheet for formulas:
> http://spreadsheets.google.com/ccc?key=reZ9v3xyJKfKhHAvs9gixqA
>
> Column F has formula which compares average data from latest week with
> average data from last week. Also, it uses coefficients to make impact of
> parameters smaller. It is not perfect, still (karma can get negative under
> some circumstances, but that would be difficult to achieve). I like, that it
> gives a smooth curve to karma and helps to highlight apps which recently
> became more popular.
>
> Implementation of this formula will require 2 weeks of daily-stats. Niels
> will start to gather those (we still need to discuss how those will be
> stored).
>
> If you have any ideas on making formula better, I will be glad to
> implement those. As soon as we have enough stats-data we would be able
> to tweak formula and see results instantly
>
> --
> Alexey Zakhlestin
> http://www.milkfarmsoft.com/
>

--
Alexey Zakhlestin
http://www.milkfarmsoft.com/

Hi Alexey,

Just wondering - do the figures in the spreadsheet come from a real
application, or are they contrived? They look deliberately contrived.

How does the curve look with the formula in the wiki if, instead of the
download totals being for the current day, they're for the previous
month? The idea of previous month or previous week is to even out the
spikes you mention, while giving the same timeframe for all apps.

Cheers,
Dave.

Alexey Zakhlestin wrote:
> Hi, everyone.
> I recently started to work with nemein and among other things working
> on application karma task from current sprint.
>
> We talked with Niels Brett a yesterday and decided, that formula from here:
> http://wiki.maemo.org/Task:Karma_for_applications wouldn't work. It is too
> rough and will give a lot of spikes. Instead, I proposed another one.
> Here is a little test-spreadsheet for formulas:
> http://spreadsheets.google.com/ccc?key=reZ9v3xyJKfKhHAvs9gixqA
>
> Column F has formula which compares average data from latest week with
> average data from last week. Also, it uses coefficients to make impact of
> parameters smaller. It is not perfect, still (karma can get negative under
> some circumstances, but that would be difficult to achieve). I like, that it
> gives a smooth curve to karma and helps to highlight apps which recently
> became more popular.
>
> Implementation of this formula will require 2 weeks of daily-stats. Niels
> will start to gather those (we still need to discuss how those will be
> stored).
>
> If you have any ideas on making formula better, I will be glad to
> implement those. As soon as we have enough stats-data we would be able
> to tweak formula and see results instantly
>

--
maemo.org docsmaster
Email: dneary@maemo.org
Jabber: bolsh@jabber.org

On Fri, May 29, 2009 at 5:03 PM, Dave Neary <dneary@maemo.org> wrote:
> Hi Alexey,
>
> Just wondering - do the figures in the spreadsheet come from a real
> application, or are they contrived? They look deliberately contrived.

we don't have any real stats, at the moment (except for the "total
number of downloads"). that's the problem

> How does the curve look with the formula in the wiki if, instead of the
> download totals being for the current day, they're for the previous
> month? The idea of previous month or previous week is to even out the
> spikes you mention, while giving the same timeframe for all apps.

Well, if I understood you correctly it should be in column G now.
Fourth chart is connected to it. Looks like cosine ;)


--
Alexey Zakhlestin
http://www.milkfarmsoft.com/
Sent from St Petersburg, Spe, Russia