11:21 a.m.
Hey guys,
I am trying to come up with a way to get Drools to detect a trend in a
dataset. The data is very basic and looks something like this:
(each row represents an object in the WM)
Time Value
10:00 1
11:00 2
12:00 3
01:00 1
02:00 3
04:00 4
05:00 5
06:00 3
I want to set up a rule that detects when there is an upward trend in the
value for 3(configurable) or more consecutive data points.
It's easy to set up a rule to detect when there is a two consecutive rows
and the 2nd one has a higher value. But how can I set up a rule to detect
when there are multiple consecutive rows each with a higher value than the
previous one?
Any ideas are welcome(even if they are not complete!).
Thanks
--
View this message in context:
http://drools.46999.n3.nabble.com/drools-fusion-Detecting-a-Trend-in-Data...
Sent from the Drools: User forum mailing list archive at Nabble.com.
11:53 a.m.
Quick ideas (not tested):
1) Detect an always increasing trend:
ensure that no previous value - possibly scoped by a sliding window - is
greater than the current
$current : MyPoint( $val : value )
not MyPoint( value > $val, this before $current ) over window:length(3)
2) A more robust approach, allowing for some noise:
create a custom accumulate function that computes a (linear) regression
and returns a
data structure with the basic output parameters (regression
coefficients, confidence intervals, etc..)
You would then constrain on that... something like
when
LinearRegressionData( k > .... ) from accumulate ( $x : MyPoints()
over window:length(3),
myCustomLinearRegressionAccumulateFunction $x ) )
then
Other options might be possible
Davide
On 11/03/2013 10:21 AM, code4dc wrote:
Hey guys,
I am trying to come up with a way to get Drools to detect a trend in a
dataset. The data is very basic and looks something like this:
(each row represents an object in the WM)
Time Value
10:00 1
11:00 2
12:00 3
01:00 1
02:00 3
04:00 4
05:00 5
06:00 3
I want to set up a rule that detects when there is an upward trend in the
value for 3(configurable) or more consecutive data points.
It's easy to set up a rule to detect when there is a two consecutive rows
and the 2nd one has a higher value. But how can I set up a rule to detect
when there are multiple consecutive rows each with a higher value than the
previous one?
Any ideas are welcome(even if they are not complete!).
Thanks
--
View this message in context:
http://drools.46999.n3.nabble.com/drools-fusion-Detecting-a-Trend-in-Data...
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
12:24 a.m.
To keep it simple,
declare Trend
List entries
int lastValue
bool hot
end
One rule creates a new Trend with a single entry if there is a new Entry:
rule newTrend
when
$e: Entry(...)
not Trend
then
create Trend with $e in entries, insert
end
Two to extend or rewind a Trend:
rule extendTrend
when
$t: Trend( $lv: lastValue )
$e: Entry( $value: value >= $lv ) // or >
then
extend Trend, modify Trend
end
rule rewindTrend
when
$t: Trend( $lv: lastValue )
$e: Entry( $value: value < $lv ) // or <=
then
clear entries, add $e, modify Trend
end
One to discover a "hot" trend:
rule discoverHotTrend
when
$t: Trend( ! $hot, entries.size >= LIMIT )
then
hot becomes true, modify Trend
end
I always opt for a simple solution so that I can understand it ;-)
-W
On 03/11/2013, Davide Sottara <dsotty(a)gmail.com> wrote:
Quick ideas (not tested):
1) Detect an always increasing trend:
ensure that no previous value - possibly scoped by a sliding window - is
greater than the current
$current : MyPoint( $val : value )
not MyPoint( value > $val, this before $current ) over window:length(3)
2) A more robust approach, allowing for some noise:
create a custom accumulate function that computes a (linear) regression
and returns a
data structure with the basic output parameters (regression
coefficients, confidence intervals, etc..)
You would then constrain on that... something like
when
LinearRegressionData( k > .... ) from accumulate ( $x : MyPoints()
over window:length(3),
myCustomLinearRegressionAccumulateFunction $x ) )
then
Other options might be possible
Davide
On 11/03/2013 10:21 AM, code4dc wrote:
> Hey guys,
>
> I am trying to come up with a way to get Drools to detect a trend in a
> dataset. The data is very basic and looks something like this:
> (each row represents an object in the WM)
> Time Value
> 10:00 1
> 11:00 2
> 12:00 3
> 01:00 1
> 02:00 3
> 04:00 4
> 05:00 5
> 06:00 3
>
> I want to set up a rule that detects when there is an upward trend in the
> value for 3(configurable) or more consecutive data points.
>
> It's easy to set up a rule to detect when there is a two consecutive rows
> and the 2nd one has a higher value. But how can I set up a rule to detect
> when there are multiple consecutive rows each with a higher value than
> the
> previous one?
>
> Any ideas are welcome(even if they are not complete!).
>
> Thanks
>
>
>
>
> --
> View this message in context:
>
http://drools.46999.n3.nabble.com/drools-fusion-Detecting-a-Trend-in-Data...
> 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
>
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
2:25 p.m.
Guys thank you so much for that quick response and fantastic solutions!!
Davide, thank you for your input! The 2nd approach would be very useful
especially if I have to go from detecting unbroken trends to detecting
statical trends!
laune, I used your approach and it gave me exactly the kind of flexibility I
needed to detect different trends. Thank you again for your help, I was
about ready to give up on this!
Eugene
--
View this message in context:
http://drools.46999.n3.nabble.com/drools-fusion-Detecting-a-Trend-in-Data...
Sent from the Drools: User forum mailing list archive at Nabble.com.
4123
days inactive
4124
days old
3 comments
3 participants
participants (3)
-
code4dc -
Davide Sottara -
Wolfgang Laun