<html>
<head>
<style><!--
.hmmessage P
{
margin:0px;
padding:0px
}
body.hmmessage
{
font-size: 10pt;
font-family:Tahoma
}
--></style></head>
<body class='hmmessage'><div dir='ltr'>
<SPAN class=hps closure_uid_ikzxel="6115">Hi</SPAN> <SPAN class=hps closure_uid_ikzxel="6116">Davide</SPAN><SPAN closure_uid_ikzxel="6117">,</SPAN><BR closure_uid_ikzxel="6158"><BR closure_uid_ikzxel="6159"><SPAN class=hps closure_uid_ikzxel="6118">interesting</SPAN> the <SPAN class=hps closure_uid_ikzxel="6119">possibility of</SPAN> <SPAN class=hps closure_uid_ikzxel="6120">inverse transformation </SPAN><SPAN class=hps closure_uid_ikzxel="6122">Decision Trees</SPAN><SPAN closure_uid_ikzxel="6123">.</SPAN> <SPAN class=hps closure_uid_ikzxel="6124">We have many</SPAN> <SPAN class=hps closure_uid_ikzxel="6125">rules that use</SPAN> <SPAN class=hps closure_uid_ikzxel="6126">DT</SPAN><SPAN closure_uid_ikzxel="6127">, so if</SPAN> <SPAN class=hps closure_uid_ikzxel="6128">this functionality</SPAN> <SPAN class=hps closure_uid_ikzxel="6129">existed, it</SPAN> <SPAN class=hps closure_uid_ikzxel="6130">could help us</SPAN> <SPAN class=hps closure_uid_ikzxel="6131">to extract some</SPAN> <SPAN class=hps closure_uid_ikzxel="6132">of the reports</SPAN> <SPAN class=hps closure_uid_ikzxel="6133">we need.</SPAN> <SPAN class=hps closure_uid_ikzxel="6134"></SPAN><BR>
<SPAN class=hps closure_uid_ikzxel="6134"></SPAN> <BR>
<SPAN class=hps closure_uid_ikzxel="6134">It is</SPAN> <SPAN class=hps closure_uid_ikzxel="6135">true that we still</SPAN> <SPAN class=hps closure_uid_ikzxel="6136">would have no</SPAN> <SPAN class=hps closure_uid_ikzxel="6137">information</SPAN> <SPAN class=hps closure_uid_ikzxel="6138">of the other rules</SPAN><SPAN closure_uid_ikzxel="6139">,</SPAN> <SPAN class=hps closure_uid_ikzxel="6140">written in</SPAN> <SPAN class=hps closure_uid_ikzxel="6141">DRL</SPAN><SPAN closure_uid_ikzxel="6142"> for example</SPAN><SPAN closure_uid_ikzxel="6143">, but it</SPAN> <SPAN class=hps closure_uid_ikzxel="6144">would help</SPAN> <SPAN class=hps closure_uid_ikzxel="6145">a lot.</SPAN><BR closure_uid_ikzxel="6160"><SPAN class=hps closure_uid_ikzxel="6146"></SPAN><BR>
<SPAN class=hps closure_uid_ikzxel="6146">Is there</SPAN> <SPAN class="hps alt-edited" closure_uid_ikzxel="6147">some place where</SPAN> <SPAN class=hps closure_uid_ikzxel="6148">we could</SPAN> <SPAN class=hps closure_uid_ikzxel="6149">vote for</SPAN> <SPAN class=hps closure_uid_ikzxel="6150">this functionality</SPAN> <SPAN class=hps closure_uid_ikzxel="6151">to be prioritized</SPAN> <SPAN class=hps closure_uid_ikzxel="6152">in the</SPAN> <SPAN class=hps closure_uid_ikzxel="6153">next</SPAN> <SPAN class=hps closure_uid_ikzxel="6154">version of</SPAN> <SPAN class=hps closure_uid_ikzxel="6155">Guvnor</SPAN><SPAN closure_uid_ikzxel="6156">?</SPAN><BR>
<SPAN closure_uid_ikzxel="6156"></SPAN> <BR>
<SPAN closure_uid_ikzxel="6156">Bests,</SPAN><BR>
<SPAN closure_uid_ikzxel="6156">Henrique</SPAN><BR>
<BR> <BR>
<DIV>
<DIV id=SkyDrivePlaceholder></DIV>> Date: Tue, 3 Jan 2012 06:55:02 -0800<BR>> From: dsotty@gmail.com<BR>> To: rules-users@lists.jboss.org<BR>> Subject: Re: [rules-users] Extracting management information from the content of the rules<BR>> <BR>> As I suspected - or feared :)<BR>> Two issues here: Drools creates an abstract model for the LHS but not for<BR>> the RHS. This mean that extracting the discount from the RHS will be doable,<BR>> but hardly elegant.<BR>> The other is that you want to meta-reason on the rules. Drools verifier<BR>> might indeed give you ideas. It is a tool to extract a meta-representation<BR>> of a rule and process that as a fact, using meta-rules.<BR>> Unfortunately, you would have to write your own analysis and reporting<BR>> meta-rules.<BR>> <BR>> Given the nature of your constraints, it seems that they would be equivalent<BR>> - or close enough - to a decision tree/table. We already generate rules from<BR>> DTs: I mentioned to manstis the possibility of supporting the inverse<BR>> transformation. It's not too hard, but will not probably be ready and<BR>> available in Guvnor before the next version. Keep chasing him, if this would<BR>> <BR>> In alternative, maybe the "Monte Carlo" approach could be used: generate a<BR>> huge amount of random objects, process them, serialize them with their<BR>> original and final price and finally use some data mining tool (R, weka,<BR>> knime, etc..).<BR>> <BR>> Other options (constraint propagation, semantic classification, etc..) might<BR>> require to modify the rules, so I'm not sure that they would be feasible at<BR>> all. We can discuss if interested, of course.<BR>> <BR>> --<BR>> View this message in context: http://drools.46999.n3.nabble.com/Re-rules-users-Extracting-management-information-from-the-content-of-the-rules-tp3625961p3629294.html<BR>> Sent from the Drools: User forum mailing list archive at Nabble.com.<BR>> _______________________________________________<BR>> rules-users mailing list<BR>> rules-users@lists.jboss.org<BR>> https://lists.jboss.org/mailman/listinfo/rules-users<BR></DIV> </div></body>
</html>