faq 274300929

Question about different score with same plan

by wdg on 2018-05-01 13:37:59

---

Hi,

When I run matsim 0.10 and 0.9, I find there is a different score with the same plan. I don't the reason.

The first below is run with matsim 0.9 and the second is run with matsim 0.10.

<person id="12052000_12052000_1">
 		<plan score="99.39563697292027" selected="yes">
 			<activity type="home" link="9720" x="455903.06702539005" y="5730103.746178518" end_time="07:19:34" >
 			</activity>
 			<leg mode="car" dep_time="07:19:34" trav_time="00:07:14">
 				<route type="links" start_link="9720" end_link="8668" trav_time="00:07:14" distance="9600.179489463851">9720 6126 6128 6130 581 579 577 575 1500 1494 1482 1480 1273 1271 1269 5585 5587 5589 5591 5593 5595 5597 5599 5601 10345 10339 10341 10343 7245 7247 7249 7251 7253 7255 6229 6230 10011 10012 10252 10239 10240 243 10236 9846 10241 10218 10219 10220 179 180 10169 3552 8013 5930 3418 7672 7670 7668 8668</route>
 			</leg>
 			<activity type="work" link="8668" x="456388.92154028354" y="5734510.189661062" end_time="14:57:15" >
 			</activity>
 			<leg mode="car" dep_time="14:57:15" trav_time="00:08:11">
 				<route type="links" start_link="8668" end_link="9720" trav_time="00:08:11" distance="9649.743628236793">8668 8669 7667 7669 7671 3411 1758 8008 8007 10183 10178 165 168 169 170 10230 10227 10235 10233 10234 10255 10101 6213 6214 6215 9804 7256 7254 7252 7250 7248 7246 10344 10342 10340 10346 5602 5600 5598 5596 5594 5592 5590 5588 5586 1268 1270 1272 1479 1481 1493 1499 574 576 9718 9720</route>
 			</leg>
 			<activity type="home" link="9720" x="455903.06702539005" y="5730103.746178518" >
 			</activity>
 		</plan>

 	</person>

<person id="12052000_12052000_1">
 		<plan score="75.39563697292027" selected="yes">
 			<activity type="home" link="9720" x="455903.06702539005" y="5730103.746178518" end_time="07:19:34" >
 			</activity>
 			<leg mode="car" dep_time="07:19:34" trav_time="00:07:14">
 				<route type="links" start_link="9720" end_link="8668" trav_time="00:07:14" distance="9600.179489463851">9720 6126 6128 6130 581 579 577 575 1500 1494 1482 1480 1273 1271 1269 5585 5587 5589 5591 5593 5595 5597 5599 5601 10345 10339 10341 10343 7245 7247 7249 7251 7253 7255 6229 6230 10011 10012 10252 10239 10240 243 10236 9846 10241 10218 10219 10220 179 180 10169 3552 8013 5930 3418 7672 7670 7668 8668</route>
 			</leg>
 			<activity type="work" link="8668" x="456388.92154028354" y="5734510.189661062" end_time="14:57:15" >
 			</activity>
 			<leg mode="car" dep_time="14:57:15" trav_time="00:08:11">
 				<route type="links" start_link="8668" end_link="9720" trav_time="00:08:11" distance="9649.743628236793">8668 8669 7667 7669 7671 3411 1758 8008 8007 10183 10178 165 168 169 170 10230 10227 10235 10233 10234 10255 10101 6213 6214 6215 9804 7256 7254 7252 7250 7248 7246 10344 10342 10340 10346 5602 5600 5598 5596 5594 5592 5590 5588 5586 1268 1270 1272 1479 1481 1493 1499 574 576 9718 9720</route>
 			</leg>
 			<activity type="home" link="9720" x="455903.06702539005" y="5730103.746178518" >
 			</activity>
 		</plan>

 	</person>

---

Comments: 2

---

Re: Question about different score with same plan

by Johan W. Joubert on 2018-05-01 14:04:49

That is quite curious. My first thought was that it is a result of different random sampling (even using the same seed), but you're quite correct, the plans seem identical. The only follow-up question I might have is that you post the two config files (or were they using the exact same config.xml file?) If they are different files, you may want to just check the scoring function parameters in the PlanCalcScore module in the config file.

---

Re: Question about different score with same plan

by Kai Nagel on 2018-05-01 15:38:35

Assuming that the mobsim does exactly the same (which is not guaranteed), there are some switches in the scoring where we changed the default settings. The ones I can remember right now are

• typicalDurationScoreComputation, where the default used to be uniform and is now relative.

• usingOldScoringBelowZeroUtilityDuration, where the default used to be true and is now false.

Could you please check if one of those might be the reason (i.e. first compare the config settings in the logfile, and then set, in 0.10.0, the config to the old setting and rerun)? If that's the reason, I could explain why we changed it.

---

EDIT: Reason. With typicalDurationScoreComputation=uniform, all scores at the typical durations are the same. That has the consequence that short activities accumulate more reward per time unit than long activities. With the result that agents have a tendency to trade long activities (home, work) for short ones (shop, dropoff). Not plausible.

We have known this for a long time and thus always said "don't change activities", but still wanted to make a step into the right direction. The new relative setting now sets the score at the typical duration such that the reward per time unit is the same for all activity types. This is probably still not perfect, but far better.

For more info, check the userguide under https://matsim.org/docs/userguide/ in sections 3.2.1.2 and 3.5 (I shuffled the sections today so you may need a new download).