Page 2 of 2

Re: New diffusion stats

Posted: Fri Sep 16, 2011 11:10 am
by R/J
chrishos04 wrote:Thank you for this great tool. It's very, very useful and it looks so cool.

But I think there could be a small bug. In my holiday on Crete in May '11 I got an exotic :note-50: G/Y note from an ATM. Unfortunately it doesn't show up in the greek graph on printer codes. It's not that important for me personally but it could perhaps show a major problem.
I think that it isn't shown because it's exotic. There seems to be an arbitrary threshold (50 notes in a month, I guess) for inclusion into the graphs (so mistakes aren't counted). You can see on the graph for :note-50:s for Y Greece that the number entered is decreasing steadily and that if either G or N printed notes drop below 50, they aren't shown in the graphs anymore.

Re: New diffusion stats

Posted: Fri Sep 16, 2011 8:04 pm
by Nerzhul
Yes, the data is heavily filtered because there's literally every combination you can imagine in the database.

There are 2 independent filters - first the system tries to find significant data combinations (i.e. tries to determine which combinations are real) and then filters the output again. The second pass is necessary because some combinations were introduced after 2002.

For example 10/E/X is a real combination but was put into circulation in mid 2011. The second filter catches the noise before 2011.

Re: New diffusion stats

Posted: Fri Sep 16, 2011 8:38 pm
by SeanW
Just had a look at this yesterday - nice!

Re: New diffusion stats

Posted: Fri Sep 16, 2011 9:35 pm
by JHN-TX
Nerzhul wrote:Yes, the data is heavily filtered because there's literally every combination you can imagine in the database.

There are 2 independent filters - first the system tries to find significant data combinations (i.e. tries to determine which combinations are real) and then filters the output again. The second pass is necessary because some combinations were introduced after 2002.

For example 10/E/X is a real combination but was put into circulation in mid 2011. The second filter catches the noise before 2011.
I only had a quick look first time round.
Did not know that you could click on each slice ,to get more detailed information.
I am even more impressed now 8O There is so much information that is easily and quickly accessed.

Re: New diffusion stats

Posted: Fri Sep 16, 2011 10:30 pm
by dserrano5
Nerzhul wrote:Yes, the data is heavily filtered because there's literally every combination you can imagine in the database.
What about a (joint) effort to do some cleanup? There's a lot of knowledge in the forum and it could be leveraged to fix many of the issues here. Besides that, some unknown combos are easily fixed, eg we could assume all 19000 P/N notes are really F/N, given how easy it is to misread the small F for a P.
Nerzhul wrote:(i.e. tries to determine which combinations are real)
Is it really needed to determine real combinations? Aren't they well known? (Honest question—I actually think they are).

Re: New diffusion stats

Posted: Sat Sep 17, 2011 9:09 am
by marty44
you can use my little tool to check your notes
http://www.arbeitsplatzvernichtung-durc ... check.html" onclick="window.open(this.href);return false;

I would also like to run it over the full EBT database ;)

Re: New diffusion stats

Posted: Sat Sep 17, 2011 1:17 pm
by JHN-TX
marty44 wrote:you can use my little tool to check your notes


I would also like to run it over the full EBT database ;)
Would be afraid to click on anyone`s little tool when I don`t know what trouble it could bring.

Re: New diffusion stats

Posted: Sat Sep 17, 2011 1:40 pm
by Nerzhul
dserrano5 wrote:
Nerzhul wrote:(i.e. tries to determine which combinations are real)
Is it really needed to determine real combinations? Aren't they well known? (Honest question—I actually think they are).
Well, there are 2 approaches to this problem - you either whitelist every known combination or try to statistically analyze the data. Both have advantages and disadvantages.

In the end I chose the algorithmic, automated approach because it requires minimal maintenance. When a new combination is brought into circulation (like said 10/E/X) nothing has to be changed, eventually the new data should simply show up. Of course this comes at the price of less accuracy.

But this can be changed later anyway, so for now let's see how the current system performs.

Re: New diffusion stats

Posted: Sat Sep 17, 2011 2:09 pm
by vermeer
Vatican city looks like a kind of painting.

Re: New diffusion stats

Posted: Sun Sep 18, 2011 2:16 pm
by JHN-TX
K014 on printer plate graph,for T notes. :?:

Re: New diffusion stats

Posted: Wed Sep 21, 2011 8:19 am
by shenglu
Very pleased to see the pie chart changes, when you move the cursor along the histogram and you can do for each denomination. It's like your own animation.That Yes, yes I am very intuitive meaning.

Re: New diffusion stats

Posted: Wed Sep 21, 2011 5:23 pm
by JHN-TX
shenglu wrote:Very pleased to see the pie chart changes, when you move the cursor along the histogram and you can do for each denomination. It's like your own animation.That Yes, yes I am very intuitive meaning.
It means that you can see in a few seconds what would have taken a lot longer before.When you get an old K/T :note-10: note you can pinpoint exactly when it was first entered.If you get a K001/T :note-10: in poor condition it is interesting to know how long it has been circulating.
EDIT
If you run the "animation" and look at :flag-be: notes for example,
46% of notes entered in Feb 2002 were :flag-be: , and this dropped to a low of
2% in May 2009.It would have taken hours to find this out the old way.
You can do the same for :flag-nl: :flag-de: or any other country and it only takes a few minutes.

Edit.
I know it is not really an animation.
These are animations http://www.funnyhub.com/animations/page ... conds.html" onclick="window.open(this.href);return false;