12:40 a.m.
Hi,
I have a strange behavior with the Drools Expert 5.5.0. After running the
main class few times in a row, the results suddenly differ from the ones
before it. That means there are somtimes 3, 4 or 5 results shown in console.
I am using a statefulknowledgesession. Inserting of all facts is done with
normal java lists. For my last type of facts I insert one, call
fireAllRules(), and retract it again. Showing in console is solved with
queries. At the end I'm calling the dispose()-method. I added it to see if
there is a change in behavior. But with or without it there is the same
problem with different results after few directly runnings.
Rules and queries are defined in a drl-file and loaded via
knowledgebuilder's add()-method. I am working with eclipse and of course the
plugin for it.
Could it be a problem with something memory like (in a way like a cache or
similar to the garbage collection)?
I hope with this short explanation you can give me a feedback. If not its
not problem to describe in a more detailed way.
Thank you.
--
View this message in context:
http://drools.46999.n3.nabble.com/Different-results-with-same-fact-base-t...
Sent from the Drools: User forum mailing list archive at Nabble.com.
6:37 a.m.
If your program terminates without any new session, dispose() need
not be called.
Getting different results for different runs with the same program+data
is strange unless you use generated random test data or rules based on
time in some way.
That's all I can say given the info provided.
-W
On 29/11/2012, raffi <christian.koehler(a)s2008.tu-chemnitz.de> wrote:
Hi,
I have a strange behavior with the Drools Expert 5.5.0. After running the
main class few times in a row, the results suddenly differ from the ones
before it. That means there are somtimes 3, 4 or 5 results shown in
console.
I am using a statefulknowledgesession. Inserting of all facts is done with
normal java lists. For my last type of facts I insert one, call
fireAllRules(), and retract it again. Showing in console is solved with
queries. At the end I'm calling the dispose()-method. I added it to see if
there is a change in behavior. But with or without it there is the same
problem with different results after few directly runnings.
Rules and queries are defined in a drl-file and loaded via
knowledgebuilder's add()-method. I am working with eclipse and of course
the
plugin for it.
Could it be a problem with something memory like (in a way like a cache or
similar to the garbage collection)?
I hope with this short explanation you can give me a feedback. If not its
not problem to describe in a more detailed way.
Thank you.
--
View this message in context:
http://drools.46999.n3.nabble.com/Different-results-with-same-fact-base-t...
Sent from the Drools: User forum mailing list archive at Nabble.com.
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
6:56 a.m.
laune wrote
If your program terminates without any new session, dispose() need
not be called.
I only have one session during every run. So using dispose is general
correct in my case.
laune wrote
Getting different results for different runs with the same
program+data
is strange unless you use generated random test data or rules based on
time in some way.
That's all I can say given the info provided.
-W
My data is stored in csv-files. I transform it via Smooks-Framework into
normal java lists. From there I insert every element. The content of the
files is master data, so there is no changing until I edit them manually.
For my rules I am only using /salience/ and /activation-group/. In the
conclusion parts I work with all 3 standard functions (insert, modify,
retract) and only with these.
I will test it on another pc to see if this could be a Drools' problem.
Thanks so far...
--
View this message in context:
http://drools.46999.n3.nabble.com/Different-results-with-same-fact-base-t...
Sent from the Drools: User forum mailing list archive at Nabble.com.
2:59 a.m.
Hi.
I localized the problem. The following lines lead to the described behavior:
/for (Translog tr : translog) {
if (tr.getTransType() == 1 && !(tr.getUserID().compareTo("null") == 0))
{
FactHandle tr_insert = ksession.insert(tr);
ksession.fireAllRules();
ksession.retract(tr_insert);
}
}/
If I run the program with the changed lines
/for (Translog tr : translog) {
if (tr.getTransType() == 1 && !(tr.getUserID().compareTo("null") == 0))
{
ksession.insert(tr);
}
}
ksession.fireAllRules();/
it doesn't work as I expect it. I didn't change the according rule
/rule "filter products with category and brand for slot 1"
salience 20
activation-group "productfilter slot 1"
when
Translog($pid : itemID, $uid : userID)
Product(pid == $pid, brand != "", $price : netUnitPrice,
$brand : brand)
ProductCategory(pid == $pid, $cid : cid, $root : root)
Category(cid == $cid, $cname : cname)
FilterCriteria(categories_slot1 contains $root)
FilterCriteria(brands contains $brand)
not Recommendation(pid == $pid, slot == 1, uid == $uid)
then
insert(new Recommendation($pid, $price, $brand, $cname, $cid, (byte)1,
$root, $uid));
end/
In the end I show all Recommendation-facts in the console using a query.
There should be at least 4, grouped by uid (userid). But with the corrected
lines I only get one Recommendation-fact.
/query "recomm for slot 1"
$recomm_max : Recommendation(slot == 1, $recomm_value : recommValue, $uid :
uid)
not Recommendation(slot == 1, recommValue > $recomm_value, uid == $uid)
end/
The same applies for the query, without grouping by uid:
/query "recomm for slot 1"
$recomm_max : Recommendation(slot == 1)
end/
I chose the first (wrong) solution, with retracting the Translog-facts
directly after firing, because it is transaction data. For my test there are
only a few lines, but in the productive scenario there are at least 100.000
records to insert.
Regards...
--
View this message in context:
http://drools.46999.n3.nabble.com/Different-results-with-same-fact-base-t...
Sent from the Drools: User forum mailing list archive at Nabble.com.
3:17 a.m.
While thinking about the problem I remembered Fusion and CEP. Could this be a
scenario for /entry-point/? The problem is, I have one rule, which minimizes
the master data. It is called before the one mentioned above with a higher
salience:
/
rule "exclude products"
salience 25
when
$p : Product(netUnitPrice < 4.00)
or
$p : Product(onlineFlag == false)
then
retract($p);
end/
I would add the /entry-point/ statement only to the above mentioned
filter-rule and change back to my first solution (inserting Translog, firing
rules, retracting Translog).
Regards...
--
View this message in context:
http://drools.46999.n3.nabble.com/Different-results-with-same-fact-base-t...
Sent from the Drools: User forum mailing list archive at Nabble.com.
4:32 a.m.
I don't see how these two loops should result in different facts in WM
unless there
are other rules besides the one you have shown.
Why do you expect more that one result from the first (!) version of
"recomm for slot 1"? Are there several recommendations with the same
maximum recommValue?
rule "filter products with category and brand for slot 1" isn't optimal.
The
patterns should be rearranged for earliest possible blocking of evaluation.
rule "filter products with category and brand for slot 1"
salience 20
activation-group "productfilter slot 1"
when
Translog($pid : itemID, $uid : userID)
Product(pid == $pid, brand != "", $price : netUnitPrice,
$brand : brand)
FilterCriteria(brands contains $brand)
ProductCategory(pid == $pid, $cid : cid, $root : root)
FilterCriteria(categories_slot1 contains $root) // stop
here if the root needs filtering
not Recommendation(pid == $pid, slot == 1, uid == $uid)
Category(cid == $cid, $cname : cname) // just needed for
the insert
then
insert(new Recommendation($pid, $price, $brand, $cname, $cid,
(byte)1, $root, $uid));
end
Regarding your rule "exclude products", it should probably be run before
you insert any
Translog facts, with a separate fireAllRules() call. Then you don't need
salience.
Or you can include the conditions in the Product pattern of
rule "filter products with category and brand for slot 1"
-W
On 3 December 2012 09:59, raffi <christian.koehler(a)s2008.tu-chemnitz.de>wrote:
Hi.
I localized the problem. The following lines lead to the described
behavior:
/for (Translog tr : translog) {
if (tr.getTransType() == 1 &&
!(tr.getUserID().compareTo("null")
== 0)) {
FactHandle tr_insert = ksession.insert(tr);
ksession.fireAllRules();
ksession.retract(tr_insert);
}
}/
If I run the program with the changed lines
/for (Translog tr : translog) {
if (tr.getTransType() == 1 &&
!(tr.getUserID().compareTo("null")
== 0)) {
ksession.insert(tr);
}
}
ksession.fireAllRules();/
it doesn't work as I expect it. I didn't change the according rule
/rule "filter products with category and brand for slot 1"
salience 20
activation-group "productfilter slot 1"
when
Translog($pid : itemID, $uid : userID)
Product(pid == $pid, brand != "", $price : netUnitPrice,
$brand : brand)
ProductCategory(pid == $pid, $cid : cid, $root : root)
Category(cid == $cid, $cname : cname)
FilterCriteria(categories_slot1 contains $root)
FilterCriteria(brands contains $brand)
not Recommendation(pid == $pid, slot == 1, uid == $uid)
then
insert(new Recommendation($pid, $price, $brand, $cname,
$cid, (byte)1,
$root, $uid));
end/
In the end I show all Recommendation-facts in the console using a query.
There should be at least 4, grouped by uid (userid). But with the corrected
lines I only get one Recommendation-fact.
/query "recomm for slot 1"
$recomm_max : Recommendation(slot == 1, $recomm_value :
recommValue, $uid :
uid)
not Recommendation(slot == 1, recommValue > $recomm_value, uid ==
$uid)
end/
The same applies for the query, without grouping by uid:
/query "recomm for slot 1"
$recomm_max : Recommendation(slot == 1)
end/
I chose the first (wrong) solution, with retracting the Translog-facts
directly after firing, because it is transaction data. For my test there
are
only a few lines, but in the productive scenario there are at least 100.000
records to insert.
Regards...
--
View this message in context:
http://drools.46999.n3.nabble.com/Different-results-with-same-fact-base-t...
Sent from the Drools: User forum mailing list archive at Nabble.com.
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
12:38 a.m.
laune wrote
I don't see how these two loops should result in different facts
in WM
unless there
are other rules besides the one you have shown.
Yes, there are other rules. The complete result should be 4 recommendations
for every customer (uid) from the translogs. The recommended products are
for a newsletter. 4 products (4 slots) should be shown there, that is why I
have the /slot == 1/ statement. For simplicity I only showed the rules for
slot 1.
So in the end there are 4 slots, all with one recommendation for every
customer. My rules for slot 1, are the following (always two, because there
are products within a category and brand and also only within a category):
/rule "exclude products"
salience 25
when
$p : Product(netUnitPrice < 4.00)
or
$p : Product(onlineFlag == false)
then
retract($p);
end/
/rule "filter products with category and brand for slot 1"
salience 20
activation-group "productfilter slot 1"
when
Translog($pid : itemID, $uid : userID)
Product(pid == $pid, brand != "", $price : netUnitPrice,
$brand : brand)
ProductCategory(pid == $pid, $cid : cid, $root : root)
Category(cid == $cid, $cname : cname)
FilterCriteria(categories_slot1 contains $root)
FilterCriteria(brands contains $brand)
not Recommendation(pid == $pid, slot == 1, uid == $uid)
then
insert(new Recommendation($pid, $price, $brand, $cname, $cid, (byte)1,
$root, $uid));
end
rule "filter products only with category for slot 1"
salience 20
activation-group "productfilter slot 1"
when
Translog($pid : itemID, $uid : userID)
Product(pid == $pid, brand == "", $price : netUnitPrice)
ProductCategory(pid == $pid, $cid : cid, $root : root)
Category(cid == $cid, $cname : cname)
FilterCriteria(categories_slot1 contains $root)
not Recommendation(pid == $pid, slot == 1, uid == $uid)
then
insert(new Recommendation($pid, $price, "", $cname, $cid, (byte)1, $root,
$uid));
end/
/rule "calculate recomm value for slot 1 with category and brand"
salience 15
activation-group "calculate slot 1"
when
$recomm : Recommendation(recommValue == 0, slot == 1, brand != "",
$price : price, $recomm_value : recommValue,
$brand : brand, $root : root)
$f : FilterCriteria()
$cat_map : HashMap() from $f.getCategories_slot1()
$brand_map : HashMap() from $f.getBrands()
then
modify ($recomm) {
setRecommValue(((Double)$cat_map.get($root)).doubleValue()
+ ((Double)$brand_map.get($brand)).doubleValue()
+ $price)
};
end
rule "calculate recomm value for slot 1 only with category"
salience 15
activation-group "calculate slot 1"
when
$recomm : Recommendation(recommValue == 0, slot == 1, brand == "",
$price : price, $recomm_value : recommValue,
$root : root)
$f : FilterCriteria()
$cat_map : HashMap() from $f.getCategories_slot1()
then
modify ($recomm) {
setRecommValue(((Double)$cat_map.get($root)).doubleValue()
+ $price)
};
end/
laune wrote
Why do you expect more that one result from the first (!) version of
"recomm for slot 1"? Are there several recommendations with the same
maximum recommValue?
Yes that is possible, but with a different /$uid/. Because there is more
than one customer, coming from the Translog-facts.
laune wrote
rule "filter products with category and brand for slot 1"
isn't optimal.
The
patterns should be rearranged for earliest possible blocking of
evaluation.
OK. That makes sense to me. Does it have a positive impact on performance of
the engine?
To call fireAllRules before inserting Translog-facts makes also sense to me,
thanks for advice.
Regards...
--
View this message in context:
http://drools.46999.n3.nabble.com/Different-results-with-same-fact-base-t...
Sent from the Drools: User forum mailing list archive at Nabble.com.
4404
days inactive
4409
days old
6 comments
2 participants
participants (2)
-
raffi -
Wolfgang Laun