| Index | Recent Threads | Who's Online | User List | Search |
|
|
 |
Portfolio123 » List all forums » Forum: Announcements » Thread: Update |
|
Total posts in this thread: 11
|
[Request a Feature] [Add To My Favorites] [Watch this Thread] |
| Author |
|
|
marco
|
I believe we found the cause of the instability in a relatively small portion of the code and made a patch. I'll turn on the sim servers to see the behavior under some prolonged stress period. I believe there's still a problem though, so I don't think it will last through the day. This is actually a big step since we were looking for the cause everywhere. On a separate note I've noticed some very large Ranking Systems that suck up resources.There are ranking systems with over 100 factors that take about 2 minutes to rank. If you have such systems please read my previous post: http://www.portfolio123.com/mvnforum/viewthread?thread=1437 Not sure what is the best way to handle these systems. At the moment there are no limits to the number of inputs in a ranking system. We may have to re-think that. A limit, even a relatively high one, is needed since people can get carried away adding everything. A system with too many factors is not effective. Factors tend to be highly correlated and more factors are simpy going to dilute the effective ones. Any suggestions or reasons why big systems are needed are appreciated. We'll also buy a few more servers and see how we can improve the ranking performance. Regards ---------------------------------------- Portfolio123 Staff |
||
|
|
azouz
Advanced Member
|
Marco, i have a ranking system that have more than 100 inputs, but all theses input have a weight of 0% and only 5 of them have a weight of 20%. Is this kind of ranking system take a lot of time for calculation ? I procced this way so i can make quickly a change if i want to choose another factor. |
||
|
|
azouz
Advanced Member
|
i am asking this question because a simulation now take about 15minutes or more to finish. When i run a simulation i get for example : Ranking started Processing 03/31/2001 04/08/2001 04/15/2001 04/21/2001 then it stay here for about 2 minutes, reloading the same date , then add one or two date and again stay some times.... |
||
|
|
markajohn
Advanced Member
|
azouz, Why not just keep the big version of the ranking system and create a new small subset of the big one by deleting what you don't want to use. That would be pretty fast and simple. Only an experiment will tell if using the smaller one is faster than the bigger 95%-uinused one. My 2ยข MJ  ---------------------------------------- [Edit 2 times, last edit by markajohn at Mar 28, 2006 8:39:34 AM] |
||
|
|
charley
Advanced Member  UNITED STATES Joined: Jul 8, 2004 Posts: 298 Status: Offline |
Not only are the ranking systems larger but it seems that more and more sophisticated functions are being developed within the ranking stock factors and Portfolio buy and sell criteria. I would suspect this would increase number crunching demand as well. Maybe this is why other commercial products are more limited in what you can do. Maybe eventually P123 may have to regulates Processor usage somehow, or invest in a Crey. One could set a limit such as # stock factors-variables used rather than absolute number of ranking system. Otherwise - P123 is doing a great job - it has been a major source of income foe me since the beta testing days. ---------------------------------------- Charley |
|||
|
|
charley
Advanced Member UNITED STATES Joined: Jul 8, 2004 Posts: 298 Status: Offline |
Correction: CRAY ---------------------------------------- Charley |
|||
|
|
jeffhansen
Member
|
Did you guys restore a DB backup as well? A sim and port I created two weeks ago is now missing.  |
||
|
|
budonk
Advanced Member
|
its hard to know how (if) factors are correlated withought being able to test it. either require a higher subscription for the high # of factors, or invest in more computing power. |
||
|
|
Barn
Advanced Member
|
I agree, if you need more computing power for the higher factor ranking systems then that cost should flow back to the people that want to use it. I wouldn't be offended if factor limits went with subscription status. ---------------------------------------- Just my $0.02 ... Actually I need that back for my port. Barn |
||
|
|
Stittsville123
Advanced Member
|
I think there is a message getting lost here. Azouz has found a way to improve the selection of stock factors within a ranking system. His method is to call up all factors that he finds interesting. Then zero the weights for individual factors when not in use. This is much more efficient than trying to cut and re-select individual stock factors. Especially when you have hundreds of factors to try in different combinations. The answer is not to prevent people from creating large ranking systems (for reason described above) but to provide a more efficient user interface. It is to P123's benefit since better ranking systems will ultimately be produced. Steve |
||
|
|
[Show Thread Printable Version] |
