BayesScheduler

[PegasusWar]

Hi,

I've developed an experimental scheduler that uses Bayes technology to "guess" what shows the user is interested in and is able to add it to gbpvr.mdb.

It learns by using the shows that are scheduled manually by the user (and the shows that are not scheduled in this time frame)...

Is anyone interested to try?

[jorm]

what is this bayes technology is there a website that you are quering against?

[jrockintuitive]

I would be very interested in this.

I would be happy to test as well.

I absolutely love GBPVR. However, I have no idea what new shows are on and if they are any good. All I do is watch the shows I record. Also, since I never watch commercials, thanks comskip, I have no idea if a mini-series is on. Which is a good thing, kind of.

If the system can record new shows I might like, and any mini-series on sci-fi, I would be a very happy camper.

[PegasusWar]

Thanks for your replies,

i would be happy to upload it somewhere. I don't seam to have access to the wiki with my forum account, do I?

[PegasusWar]

Thanks for your help.

[PegasusWar]

OK. In opera I cannot see the File Attachments option. In IE it works...

I have extracted the zip in my gbpvr folder. It will create a folder named bayesscheduler. To add the scheduler to your post epg update process, add the line

%~dp0bayesscheduler\start.cmd >bayesscheduler.log

to the PostUpdateEPG.bat batch file.

The program is coded in java (i use 1.5 (aka 5.0) should work with 1.4).

It creates the statistic file in the gbpvr folder (named corpus.txt). You can change the settings in the start.cmd batch file. Detailed command line option are displayed if you start the batchfile with -?:

<table border="0" align="center" width="95%" cellpadding="0" cellspacing="0"><tr><td>Code Sample </td></tr><tr><td id="CODE">
D:\Programme\devnz\gbpvr\bayesscheduler>start.cmd -?
usage: BayesFilter [-s showid] [-b double value] -g gbpvrpath [--record]
      -c corpuspath [-t className] [-?]
   --record                      Add result to recordings
-?,--help                        print out this help text
-b,--borderline <double value>   border between good and bad (default:
                                 0.96)
-c,--corpus <corpuspath>         Path and filename of bayes corpus
-g,--gbpvrpath <gbpvrpath>       Path to gbpvr directory
-s,--showid <showid>             ID of show to learn
-t,--type <className>            target type for show (BAD, GOOD)

(c) M. Baeker, Revision: $Rev: 784 $, Date: $LastChangedDate: 2005-03-20
13:13:06 +0100 (So, 20 Mrz 2005) $
[/QUOTE]

[jrockintuitive]

Where is the program getting the information to make recommendation - or to record.

Do I need java installed? I can't play with it today, but hopefully later this week.

[KingArgyle]

Yeah, you'll need at least a 1.4 or 1.5(beta) JRE or SDK installed.

[jorm]

Quote[/b] ]Where is the program getting the information to make recommendation - or to record.

My question exactly. I went to the website that tipstir mentioned nothing there about tv guide. I am assuming that there is a website somewhere which has the association table.

[PegasusWar]

bayes is simple statistic. (someone seams to call it ai, but in my option it is fare away...)
To train itself it uses the gbpvr recording table. Every recording will be used to train as &quot;good shows&quot;... (by using channel name, title and description)

Every show in the same time slot will be used to train the &quot;bad shows&quot; (by using the same infos).
(it was your decision to record this show in this time slot and not the other posibilities)

in corpus.txt every word that is good or bad will be counted and increased. By calculating the propability for every word of a show in the future and combine them (Bayesian probability) you get a probability for the show. If it is above the &quot;border line&quot; it will be recorded.

it is nessasary to train the filter. (like a spam filter) so you need to record shows on your own. If you have the programm_oid you can also add it manually to the filter (as bad or good shows).