{ "id": 77700, "title": "Faults are in pairs ", "url": "/competitions/vsb-power-line-fault-detection/discussion/77700", "author_name": "", "post_date": "2019-01-15T19:44:05.017429700Z", "votes": null, "comment_count": 7, "views": 0, "content": "

Faults are mostly in pair of three \nWhy is that ?

", "messages": [ { "id": "456424", "postDate": "01/15/2019 19:44:05", "content": "

Faults are mostly in pair of three \nWhy is that ?

", "rawMarkdown": "Faults are mostly in pair of three \nWhy is that ?", "votes": null }, { "id": "458464", "postDate": "01/19/2019 17:55:06", "content": "

I believe the classification target in the training data is tied to the measurement_id which includes three phases (3 signal ids). Presumably, even if just one signal in the set of three is faulty, they mark the entire measurement of 3 signals faulty.

\n\n

This assumption of mine seems somewhat disconnected from the format for submission where they intend on submitting a classification on a per signal basis, not per measurement.

", "rawMarkdown": "I believe the classification target in the training data is tied to the measurement_id which includes three phases (3 signal ids). Presumably, even if just one signal in the set of three is faulty, they mark the entire measurement of 3 signals faulty.\n\nThis assumption of mine seems somewhat disconnected from the format for submission where they intend on submitting a classification on a per signal basis, not per measurement.", "votes": null }, { "id": "458563", "postDate": "01/19/2019 23:37:00", "content": "

I counted the numbers of faulted signals per measurement in metadata_train.csv: 2710 have no faults, 19 have 1, 19 have 2, and 156 have all 3.

", "rawMarkdown": "I counted the numbers of faulted signals per measurement in metadata_train.csv: 2710 have no faults, 19 have 1, 19 have 2, and 156 have all 3.", "votes": null }, { "id": "458943", "postDate": "01/20/2019 21:15:50", "content": "

One possible explanation: most of time (in real life) if someting bad happens, then it happens with all 3 phases. :)\nOr maybe this is connected to the marking process as Jeffrey Egan wrote.

", "rawMarkdown": "One possible explanation: most of time (in real life) if someting bad happens, then it happens with all 3 phases. :)\nOr maybe this is connected to the marking process as Jeffrey Egan wrote.", "votes": null }, { "id": "458952", "postDate": "01/20/2019 21:54:17", "content": "

Thanks for digging into the metadata and clarifying this! Interesting that in measurements with faults it's likely to have all three phases exhibit faults but that it is not always the case.

", "rawMarkdown": "Thanks for digging into the metadata and clarifying this! Interesting that in measurements with faults it's likely to have all three phases exhibit faults but that it is not always the case.", "votes": null }, { "id": "459079", "postDate": "01/21/2019 05:58:31", "content": "

In case you're interested, here's how to generate that table using an old-fashioned Unix/Linux shell script:

\n\n
cat metadata_train.csv | sed 1d | tr ',' ' ' | awk '\n\n(NR % 3) == 1 {\n\n    Targ1 = $4\n\n    next\n\n}\n\n(NR % 3) == 2 {\n\n    Targ2 = $4\n\n    next\n\n}\n\n{\n\n    Targ3 = $4\n\n    print Targ1 + Targ2 + Targ3\n\n}\n\n' | sort | uniq -c\n
", "rawMarkdown": "In case you're interested, here's how to generate that table using an old-fashioned Unix/Linux shell script:\n\n\tcat metadata_train.csv | sed 1d | tr ',' ' ' | awk '\n\n\t(NR % 3) == 1 {\n\n\t\tTarg1 = $4\n\n\t\tnext\n\n\t}\n\n\t(NR % 3) == 2 {\n\n\t\tTarg2 = $4\n\n\t\tnext\n\n\t}\n\n\t{\n\n\t\tTarg3 = $4\n\n\t\tprint Targ1 + Targ2 + Targ3\n\n\t}\n\n\t' | sort | uniq -c", "votes": null }, { "id": "459181", "postDate": "01/21/2019 11:03:02", "content": "

It seems PD signals in multiple phases are common, at least in the lab setting. The signal intensities vary on different phases and voltages of the transimission line at least. \nYou can refer to this paper http://iopscience.iop.org/article/10.1088/1757-899X/294/1/012076/pdf

\n\n

And this is a nice introduction of PD on line survey http://www.epowerplus.com/cm/dpl/downloads/content/112/IQ_Partial_Discharge_Presentation.ppt

", "rawMarkdown": "It seems PD signals in multiple phases are common, at least in the lab setting. The signal intensities vary on different phases and voltages of the transimission line at least. \nYou can refer to this paper http://iopscience.iop.org/article/10.1088/1757-899X/294/1/012076/pdf\n\nAnd this is a nice introduction of PD on line survey http://www.epowerplus.com/cm/dpl/downloads/content/112/IQ_Partial_Discharge_Presentation.ppt", "votes": null }, { "id": "459251", "postDate": "01/21/2019 12:57:22", "content": "

Thanks :)

", "rawMarkdown": "Thanks :)", "votes": null } ], "comments": [ { "id": 458464, "author_name": "jeffreyegan", "author_url": "", "post_date": "01/19/2019 17:55:06", "content": "

I believe the classification target in the training data is tied to the measurement_id which includes three phases (3 signal ids). Presumably, even if just one signal in the set of three is faulty, they mark the entire measurement of 3 signals faulty.

\n\n

This assumption of mine seems somewhat disconnected from the format for submission where they intend on submitting a classification on a per signal basis, not per measurement.

", "votes": null, "replies": [ { "id": 458563, "author_name": "dslate", "author_url": "", "post_date": "01/19/2019 23:37:00", "content": "

I counted the numbers of faulted signals per measurement in metadata_train.csv: 2710 have no faults, 19 have 1, 19 have 2, and 156 have all 3.

", "votes": null, "replies": [] }, { "id": 458952, "author_name": "jeffreyegan", "author_url": "", "post_date": "01/20/2019 21:54:17", "content": "

Thanks for digging into the metadata and clarifying this! Interesting that in measurements with faults it's likely to have all three phases exhibit faults but that it is not always the case.

", "votes": null, "replies": [] }, { "id": 459079, "author_name": "dslate", "author_url": "", "post_date": "01/21/2019 05:58:31", "content": "

In case you're interested, here's how to generate that table using an old-fashioned Unix/Linux shell script:

\n\n
cat metadata_train.csv | sed 1d | tr ',' ' ' | awk '\n\n(NR % 3) == 1 {\n\n    Targ1 = $4\n\n    next\n\n}\n\n(NR % 3) == 2 {\n\n    Targ2 = $4\n\n    next\n\n}\n\n{\n\n    Targ3 = $4\n\n    print Targ1 + Targ2 + Targ3\n\n}\n\n' | sort | uniq -c\n
", "votes": null, "replies": [] } ] }, { "id": 458943, "author_name": "sergeyzlobin", "author_url": "", "post_date": "01/20/2019 21:15:50", "content": "

One possible explanation: most of time (in real life) if someting bad happens, then it happens with all 3 phases. :)\nOr maybe this is connected to the marking process as Jeffrey Egan wrote.

", "votes": null, "replies": [] }, { "id": 459181, "author_name": "steinhafen", "author_url": "", "post_date": "01/21/2019 11:03:02", "content": "

It seems PD signals in multiple phases are common, at least in the lab setting. The signal intensities vary on different phases and voltages of the transimission line at least. \nYou can refer to this paper http://iopscience.iop.org/article/10.1088/1757-899X/294/1/012076/pdf

\n\n

And this is a nice introduction of PD on line survey http://www.epowerplus.com/cm/dpl/downloads/content/112/IQ_Partial_Discharge_Presentation.ppt

", "votes": null, "replies": [ { "id": 459251, "author_name": "mks2192", "author_url": "", "post_date": "01/21/2019 12:57:22", "content": "

Thanks :)

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "456424": "Faults are mostly in pair of three \nWhy is that ?", "458464": "I believe the classification target in the training data is tied to the measurement_id which includes three phases (3 signal ids). Presumably, even if just one signal in the set of three is faulty, they mark the entire measurement of 3 signals faulty.\n\nThis assumption of mine seems somewhat disconnected from the format for submission where they intend on submitting a classification on a per signal basis, not per measurement.", "458563": "I counted the numbers of faulted signals per measurement in metadata_train.csv: 2710 have no faults, 19 have 1, 19 have 2, and 156 have all 3.", "458943": "One possible explanation: most of time (in real life) if someting bad happens, then it happens with all 3 phases. :)\nOr maybe this is connected to the marking process as Jeffrey Egan wrote.", "458952": "Thanks for digging into the metadata and clarifying this! Interesting that in measurements with faults it's likely to have all three phases exhibit faults but that it is not always the case.", "459079": "In case you're interested, here's how to generate that table using an old-fashioned Unix/Linux shell script:\n\n\tcat metadata_train.csv | sed 1d | tr ',' ' ' | awk '\n\n\t(NR % 3) == 1 {\n\n\t\tTarg1 = $4\n\n\t\tnext\n\n\t}\n\n\t(NR % 3) == 2 {\n\n\t\tTarg2 = $4\n\n\t\tnext\n\n\t}\n\n\t{\n\n\t\tTarg3 = $4\n\n\t\tprint Targ1 + Targ2 + Targ3\n\n\t}\n\n\t' | sort | uniq -c", "459181": "It seems PD signals in multiple phases are common, at least in the lab setting. The signal intensities vary on different phases and voltages of the transimission line at least. \nYou can refer to this paper http://iopscience.iop.org/article/10.1088/1757-899X/294/1/012076/pdf\n\nAnd this is a nice introduction of PD on line survey http://www.epowerplus.com/cm/dpl/downloads/content/112/IQ_Partial_Discharge_Presentation.ppt", "459251": "Thanks :)" }, "source": "meta" }