|
|
Post by nwilliard on Jul 1, 2015 19:23:21 GMT
Score is as follows.
• 4 points for every correct detection of a fault. The means that the fault code must be correct & the start time must be within + or – 1 hour of the actual start time & the end time must be within + or – the actual end time. (true positive)
• - 1 point for every miss classification. That means that the start and end times are correct within + or – 1 hour however the fault code is not correct (miss-classification)
• -10 points for every false alarm. That means when the model predicts an alarm but there is no actual fault (false positive)
• -1 point for ever missed alarm. That means a fault occurs but the model does not identify it. (false negative)
Index where first fault is taken away. All the data prior to this index is fully in tact. After this index random faults are removed.
plant_41c.csv 1023
plant_42c.csv 1882
plant_43c.csv 588
plant_45c.csv 811
plant_46c.csv 520
plant_47c.csv 913
plant_48c.csv 5794
plant_49c.csv 1429
plant_53c.csv 879
plant_54c.csv 420
plant_56c.csv 906
plant_57c.csv 1145
plant_58c.csv 1796
plant_59c.csv 1922
plant_60c.csv 1715
|
|
|
|
Post by MSA on Jul 6, 2015 15:34:39 GMT
I am trying to find the missing faults at the indexes you have mentioned. It seems like nothing is missing or removed in those locations or in any other indexes after them. Am I missing something?
|
|
cyang
New Member
Posts: 1 
|
Post by cyang on Jul 6, 2015 19:22:54 GMT
Yes! I had the same problem. The data after given index for each file is still there.
|
|
|
|
Post by nwilliard on Jul 21, 2015 21:14:17 GMT
You are not supposed to know what faults are missing. The missing faults are the faults that YOU need to predict. All the the lines up to the index mentioned in the first post are complete, there is no data missing. But after that, random lines are removed and you must identify where these missing faults are.
|
|
|
|
Post by Sam on Jul 27, 2015 16:45:08 GMT
I have the same problem here. I don't see any removed lines after those indexes. Can you please explain, for example what happens at index 1023 of plant_41c.csv? I can see the fault code 1 at that index. What are we supposed to predict?
|
|
|
|
Post by justinian on Jul 27, 2015 20:28:58 GMT
Index 1023 of plant_41c.csv shows the following true fault:
11/19/2011 9:59 11/19/2011 12:59 1
Beyond this index (that is beyond the time corresponding to this index, 11/19/2011 9:59) some faults with fault code 1-5 have been randomly removed. You cannot see these faults. Faults still visible after index 1023 are true faults as well, but their set is incomplete. The task is to predict faults beyond index 1023 aside from the ones included in the fault file.
|
|
|
|
Post by BUAA-RMS on Jul 28, 2015 2:32:17 GMT
We have questions that should the time gap of predicted fault is exactly what it is in test files. For example,the 1024 line of plant_41c.csv is as follows 2011/11/19 17:44 2011/11/19 22:14 1
From 17:44 to 22:14 it contains 5 hours, but the sampling interval of file plant_41a and plant_41b is 15 minutes.
So can we predict the fault accross 15 mins as
2011/11/19 17:44 2011/11/19 18:00 1 .?
or we must predict the fault where the time follows the test file as the boundary as
2011/11/19 17:44 2011/11/19 22:14 1. ?
|
|
|
|
Post by nwilliard on Jul 31, 2015 19:21:53 GMT
You have a + or - one hour fault tolerance on the start and end time of the faults. In order to correctly predict the fault, you must predict the start time within + or - 1 hour of the fault and also predict the end time within + or - 1 hour of the fault.
|
|
|
|
Post by BUAA-RMS on Aug 1, 2015 8:55:11 GMT
Yes,I want to know how the scroing program treat the individual fault. For example, if the real fault is as:
2011/11/19 17:44 2011/11/19 22:14 1,
can we predict as several individual faults as
2011/11/19 17:44 2011/11/19 18:00 1,
2011/11/19 18:00 2011/11/19 18:15 1,
2011/11/19 18:15 2011/11/19 18:30 1,
...
2011/11/19 22:00 2011/11/19 22:14 1,
or if wee predict as above, is it correct?
|
|
|
|
Post by nwilliard on Aug 3, 2015 5:05:17 GMT
the real fault is
2011-11-19 17:44:00 2011-11-19 22:14:00 1,
if you predict
2011-11-19 18:00:00 2011-11-19 21:16:00 1,
that would be a true positive...
if you predict
2011-11-19 18:00:00 2011-11-19 21:16:00 2,
that would be a miss classification
if you predict
2011-11-19 18:00:00 2011-11-19 21:12:00 1,
that would be a false positive.
if none of your alarms account for the real fault, it will be counted as a false negative (or a missed alarm).
|
|
|
|
Post by BUAA-RMS on Aug 3, 2015 10:01:21 GMT
So can we predict according to the time gap in all c files?
|
|
|
|
Post by wexiao on Aug 11, 2015 16:39:19 GMT
I have a question regarding false negative count of the score. I know that "if none of my alarms account for the real fault, it will be counted as a false negative". But what does none of my alarms account for the real fault mean? Does it mean none of my alarm have predict correctly (within one hour) of both the start time and end time? Or does it mean none of my alarm have predict correctly (within one hour) of the start time? Please clarify.
|
|
|
|
Post by nwilliard on Aug 12, 2015 14:35:00 GMT
both the start and the end time
|
|
|
|
Post by wexiao on Aug 12, 2015 15:04:40 GMT
So if one of the alarms has predicted correctly the start time and end time, the corresponding fault is not account for false negative. Does it need the fault type to be correct?
|
|
|
|
Post by nwilliard on Aug 20, 2015 2:16:32 GMT
you need to predict the start time within +/- one hour, you need to predict the end time within +/- one hour, and you need to have the correct fault code in order for it to count as a True Positive and hence not count as a false negative...
|
|
