\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4227513%2F5850a01b0c2831b6778f1f4aa75c552d%2F__results___25_0.png?generation=1607367549139328&alt=media\")
\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4227513%2Fa0fb12e5b5281c05be2e898cd874fe92%2F__results___25_1.png?generation=1607367574032376&alt=media\")
\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4227513%2F2d35144052f63ecd592572378c554b88%2F__results___28_0.png?generation=1607367514274035&alt=media\")
While doing preliminary analysis on the dataset, I found out this. First, have a look.
\n
\n
\n
All those class imbalance for species_id 23 and stuff are fine, but what is songtype_id? It clearly has 2 values, and type 1 dominates. In fact, no species do have type-4 songs only except 16, 17, and 23. That too for 16, there is only song type 4 is there and for 23, it's evenly distributed. Any idea what can be the take away from this?
", "messages": [ { "id": "1100217", "postDate": "12/02/2020 23:12:01", "content": "While doing preliminary analysis on the dataset, I found out this. First, have a look.
\n
\n
\n
All those class imbalance for species_id 23 and stuff are fine, but what is songtype_id? It clearly has 2 values, and type 1 dominates. In fact, no species do have type-4 songs only except 16, 17, and 23. That too for 16, there is only song type 4 is there and for 23, it's evenly distributed. Any idea what can be the take away from this?
", "rawMarkdown": "While doing preliminary analysis on the dataset, I found out this. First, have a look.\n\n\n\n\nAll those class imbalance for species_id 23 and stuff are fine, but what is songtype_id? It clearly has 2 values, and type 1 dominates. In fact, no species do have type-4 songs only except 16, 17, and 23. That too for 16, there is only song type 4 is there and for 23, it's evenly distributed. Any idea what can be the take away from this?", "votes": null }, { "id": "1101010", "postDate": "12/03/2020 14:49:51", "content": "We could make a couple extra classes and use for train and infer, then reduce for submission
\nby taking max for 17 and 23.
\nspecies_id == 17 & songtype_id == 4
\nspecies_id == 23 & songtype_id == 4
I don't know yet how this will affect the scores.
", "rawMarkdown": "We could make a couple extra classes and use for train and infer, then reduce for submission\nby taking max for 17 and 23. \nspecies_id == 17 & songtype_id == 4\nspecies_id == 23 & songtype_id == 4\n\nI don't know yet how this will affect the scores.", "votes": null }, { "id": "1102309", "postDate": "12/04/2020 19:21:09", "content": "A bird may sing multiple songs. So it's the same species, but a different song for that species. As Thomas suggested, you could make a model that predicts Species+Song, then use logic to reduce it to just Species.
\nSong ID is just an identifier, so the number 1 vs 4 doesn't mean anything to us other than representing a different song for each species.
", "rawMarkdown": "A bird may sing multiple songs. So it's the same species, but a different song for that species. As Thomas suggested, you could make a model that predicts Species+Song, then use logic to reduce it to just Species.\n\nSong ID is just an identifier, so the number 1 vs 4 doesn't mean anything to us other than representing a different song for each species.", "votes": null } ], "comments": [ { "id": 1101010, "author_name": "tomkkk", "author_url": "", "post_date": "12/03/2020 14:49:51", "content": "We could make a couple extra classes and use for train and infer, then reduce for submission
\nby taking max for 17 and 23.
\nspecies_id == 17 & songtype_id == 4
\nspecies_id == 23 & songtype_id == 4
I don't know yet how this will affect the scores.
", "votes": null, "replies": [] }, { "id": 1102309, "author_name": "maltonji", "author_url": "", "post_date": "12/04/2020 19:21:09", "content": "A bird may sing multiple songs. So it's the same species, but a different song for that species. As Thomas suggested, you could make a model that predicts Species+Song, then use logic to reduce it to just Species.
\nSong ID is just an identifier, so the number 1 vs 4 doesn't mean anything to us other than representing a different song for each species.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1100217": "While doing preliminary analysis on the dataset, I found out this. First, have a look.\n\n\n\n\nAll those class imbalance for species_id 23 and stuff are fine, but what is songtype_id? It clearly has 2 values, and type 1 dominates. In fact, no species do have type-4 songs only except 16, 17, and 23. That too for 16, there is only song type 4 is there and for 23, it's evenly distributed. Any idea what can be the take away from this?", "1101010": "We could make a couple extra classes and use for train and infer, then reduce for submission\nby taking max for 17 and 23. \nspecies_id == 17 & songtype_id == 4\nspecies_id == 23 & songtype_id == 4\n\nI don't know yet how this will affect the scores.", "1102309": "A bird may sing multiple songs. So it's the same species, but a different song for that species. As Thomas suggested, you could make a model that predicts Species+Song, then use logic to reduce it to just Species.\n\nSong ID is just an identifier, so the number 1 vs 4 doesn't mean anything to us other than representing a different song for each species." }, "source": "meta" }