I've attached an example from Subject One in which the event was categorized as multiple event-types.
\n\nHow is that possible? Is that "wrong" -- should we be removing / cleansing stuff like that from the training?
\n\n(It seems wrong to me)
", "messages": [ { "id": "88085", "postDate": "08/03/2015 13:01:28", "content": "I've attached an example from Subject One in which the event was categorized as multiple event-types.
\n\nHow is that possible? Is that "wrong" -- should we be removing / cleansing stuff like that from the training?
\n\n(It seems wrong to me)
", "rawMarkdown": "I've attached an example from Subject One in which the event was categorized as multiple event-types.\r\n\r\nHow is that possible? Is that \"wrong\" -- should we be removing / cleansing stuff like that from the training? \r\n\r\n(It seems wrong to me)", "votes": null }, { "id": "88092", "postDate": "08/03/2015 13:51:59", "content": "Example not attached
", "rawMarkdown": "Example not attached", "votes": null }, { "id": "88093", "postDate": "08/03/2015 13:53:37", "content": "Sorry. I had attached it. Maybe there was an issue with the upload or something. Thanks for letting me know.
", "rawMarkdown": "Sorry. I had attached it. Maybe there was an issue with the upload or something. Thanks for letting me know.", "votes": null }, { "id": "88095", "postDate": "08/03/2015 14:14:59", "content": "This probably should have been titled, "[Event] Observations with Multiple Categories".
\n\nJust curious what everyone else thinks of a single observation being categorized as two (or three) different event types.
", "rawMarkdown": "This probably should have been titled, \"[Event] Observations with Multiple Categories\". \r\n\r\nJust curious what everyone else thinks of a single observation being categorized as two (or three) different event types.", "votes": null }, { "id": "88100", "postDate": "08/03/2015 15:09:35", "content": "As you can read in the description, "The six label columns are either zero or one, depending on whether the corresponding event has occurred within ±150ms (±75frames)", so if the events are really close to each other, you will have frames with ones in more than one category. It actually happens quite often for events FirstDigitTouch, BothStartLoadPhase and LiftOff. For some trials mean distance between start of FirstDigitTouch and start of BothStartLoadPhase is below 30 frames, so you will get a lot of situations like this.
", "rawMarkdown": "As you can read in the description, \"The six label columns are either zero or one, depending on whether the corresponding event has occurred within ±150ms (±75frames)\", so if the events are really close to each other, you will have frames with ones in more than one category. It actually happens quite often for events FirstDigitTouch, BothStartLoadPhase and LiftOff. For some trials mean distance between start of FirstDigitTouch and start of BothStartLoadPhase is below 30 frames, so you will get a lot of situations like this.", "votes": null }, { "id": "88102", "postDate": "08/03/2015 15:12:11", "content": "That's where I was seeing it most frequently too.
\n\nDo you think it should be corrected for?
\n\n(And thanks!)
", "rawMarkdown": "That's where I was seeing it most frequently too.\r\n\r\nDo you think it should be corrected for? \r\n\r\n(And thanks!)", "votes": null }, { "id": "88109", "postDate": "08/03/2015 16:09:56", "content": "Well, I'm not sure what do you mean by "correct for". Definitely its not a mistake, but it may be beneficial to use this information in your model. In my model I'm not using the fact that events coincide per se, but somehow I'm trying to model fact that they are always in the same order and mean distances between them(though not as succesfully as I would like). For some events it works, for some it doesn't - standard deviations of distances between BothReleased and next HandStart and between LiftOff and Replace are so big that it is hard to do anything with it.
", "rawMarkdown": "Well, I'm not sure what do you mean by \"correct for\". Definitely its not a mistake, but it may be beneficial to use this information in your model. In my model I'm not using the fact that events coincide per se, but somehow I'm trying to model fact that they are always in the same order and mean distances between them(though not as succesfully as I would like). For some events it works, for some it doesn't - standard deviations of distances between BothReleased and next HandStart and between LiftOff and Replace are so big that it is hard to do anything with it.", "votes": null }, { "id": "88126", "postDate": "08/03/2015 18:11:24", "content": "I guess I see it as a mistake...that the experimenter couldn't react fast enough to properly tag the event as a specific type.
\n\nYour approach sounds interesting and certainly doesn't make the assumption I'm making. heh.
\n\nMy self-brewed models don't seem to score over .82 -- the LB model I just ran to see what the difference was in the output predictions (download the finished file to compare with my files) .
", "rawMarkdown": "I guess I see it as a mistake...that the experimenter couldn't react fast enough to properly tag the event as a specific type.\r\n\r\nYour approach sounds interesting and certainly doesn't make the assumption I'm making. heh. \r\n\r\nMy self-brewed models don't seem to score over .82 -- the LB model I just ran to see what the difference was in the output predictions (download the finished file to compare with my files) .", "votes": null } ], "comments": [ { "id": 88092, "author_name": "domcastro", "author_url": "", "post_date": "08/03/2015 13:51:59", "content": "Example not attached
", "votes": null, "replies": [] }, { "id": 88093, "author_name": "messels", "author_url": "", "post_date": "08/03/2015 13:53:37", "content": "Sorry. I had attached it. Maybe there was an issue with the upload or something. Thanks for letting me know.
", "votes": null, "replies": [] }, { "id": 88095, "author_name": "messels", "author_url": "", "post_date": "08/03/2015 14:14:59", "content": "This probably should have been titled, "[Event] Observations with Multiple Categories".
\n\nJust curious what everyone else thinks of a single observation being categorized as two (or three) different event types.
", "votes": null, "replies": [] }, { "id": 88100, "author_name": "", "author_url": "", "post_date": "08/03/2015 15:09:35", "content": "As you can read in the description, "The six label columns are either zero or one, depending on whether the corresponding event has occurred within ±150ms (±75frames)", so if the events are really close to each other, you will have frames with ones in more than one category. It actually happens quite often for events FirstDigitTouch, BothStartLoadPhase and LiftOff. For some trials mean distance between start of FirstDigitTouch and start of BothStartLoadPhase is below 30 frames, so you will get a lot of situations like this.
", "votes": null, "replies": [] }, { "id": 88102, "author_name": "messels", "author_url": "", "post_date": "08/03/2015 15:12:11", "content": "That's where I was seeing it most frequently too.
\n\nDo you think it should be corrected for?
\n\n(And thanks!)
", "votes": null, "replies": [] }, { "id": 88109, "author_name": "", "author_url": "", "post_date": "08/03/2015 16:09:56", "content": "Well, I'm not sure what do you mean by "correct for". Definitely its not a mistake, but it may be beneficial to use this information in your model. In my model I'm not using the fact that events coincide per se, but somehow I'm trying to model fact that they are always in the same order and mean distances between them(though not as succesfully as I would like). For some events it works, for some it doesn't - standard deviations of distances between BothReleased and next HandStart and between LiftOff and Replace are so big that it is hard to do anything with it.
", "votes": null, "replies": [] }, { "id": 88126, "author_name": "messels", "author_url": "", "post_date": "08/03/2015 18:11:24", "content": "I guess I see it as a mistake...that the experimenter couldn't react fast enough to properly tag the event as a specific type.
\n\nYour approach sounds interesting and certainly doesn't make the assumption I'm making. heh.
\n\nMy self-brewed models don't seem to score over .82 -- the LB model I just ran to see what the difference was in the output predictions (download the finished file to compare with my files) .
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "88085": "I've attached an example from Subject One in which the event was categorized as multiple event-types.\r\n\r\nHow is that possible? Is that \"wrong\" -- should we be removing / cleansing stuff like that from the training? \r\n\r\n(It seems wrong to me)", "88092": "Example not attached", "88093": "Sorry. I had attached it. Maybe there was an issue with the upload or something. Thanks for letting me know.", "88095": "This probably should have been titled, \"[Event] Observations with Multiple Categories\". \r\n\r\nJust curious what everyone else thinks of a single observation being categorized as two (or three) different event types.", "88100": "As you can read in the description, \"The six label columns are either zero or one, depending on whether the corresponding event has occurred within ±150ms (±75frames)\", so if the events are really close to each other, you will have frames with ones in more than one category. It actually happens quite often for events FirstDigitTouch, BothStartLoadPhase and LiftOff. For some trials mean distance between start of FirstDigitTouch and start of BothStartLoadPhase is below 30 frames, so you will get a lot of situations like this.", "88102": "That's where I was seeing it most frequently too.\r\n\r\nDo you think it should be corrected for? \r\n\r\n(And thanks!)", "88109": "Well, I'm not sure what do you mean by \"correct for\". Definitely its not a mistake, but it may be beneficial to use this information in your model. In my model I'm not using the fact that events coincide per se, but somehow I'm trying to model fact that they are always in the same order and mean distances between them(though not as succesfully as I would like). For some events it works, for some it doesn't - standard deviations of distances between BothReleased and next HandStart and between LiftOff and Replace are so big that it is hard to do anything with it.", "88126": "I guess I see it as a mistake...that the experimenter couldn't react fast enough to properly tag the event as a specific type.\r\n\r\nYour approach sounds interesting and certainly doesn't make the assumption I'm making. heh. \r\n\r\nMy self-brewed models don't seem to score over .82 -- the LB model I just ran to see what the difference was in the output predictions (download the finished file to compare with my files) ." }, "source": "meta" }