![\"creeper\"](\"https://i.imgur.com/x7a759q.png\")
It may be possible to improve submissions by looking at the combination of birds predicted within a single soundscape. The diverse ecosystems of Hawaii lead to the separation of many species due to their preferred habitats. I lived there for 3 years and saw most every region of every island during my time there - with the exception of Ni'ihau which is exclusive to native Hawaiians. There are jungles, deserts, dry forests, swamps, lava fields, ranch land, snow-covered mountains, and of course, beaches. It makes sense that many bird species would stay within a range of ecosystems.
\nOne example: here is a nice storm petrel (barpet). He lives in coastal areas close to the sea.
\n
And here is a Puaiohi (puaioh). This little lady is found only on Kauai in remote forests above 4000 ft.
\n![\"p\"](\"https://i.imgur.com/p9VuNy9.png\")
Alas, they will probably never meet. So if your model predicts both of these birds in the same soundscape along with others, you may be able to apply logic and rule out unlikely combinations.
", "messages": [ { "id": "1692609", "postDate": "02/16/2022 06:22:27", "content": "It may be possible to improve submissions by looking at the combination of birds predicted within a single soundscape. The diverse ecosystems of Hawaii lead to the separation of many species due to their preferred habitats. I lived there for 3 years and saw most every region of every island during my time there - with the exception of Ni'ihau which is exclusive to native Hawaiians. There are jungles, deserts, dry forests, swamps, lava fields, ranch land, snow-covered mountains, and of course, beaches. It makes sense that many bird species would stay within a range of ecosystems.
\nOne example: here is a nice storm petrel (barpet). He lives in coastal areas close to the sea.
\n
And here is a Puaiohi (puaioh). This little lady is found only on Kauai in remote forests above 4000 ft.
\n
Alas, they will probably never meet. So if your model predicts both of these birds in the same soundscape along with others, you may be able to apply logic and rule out unlikely combinations.
", "rawMarkdown": "It may be possible to improve submissions by looking at the combination of birds predicted within a single soundscape. The diverse ecosystems of Hawaii lead to the separation of many species due to their preferred habitats. I lived there for 3 years and saw most every region of every island during my time there - with the exception of Ni'ihau which is exclusive to native Hawaiians. There are jungles, deserts, dry forests, swamps, lava fields, ranch land, snow-covered mountains, and of course, beaches. It makes sense that many bird species would stay within a range of ecosystems. \n\nOne example: here is a nice storm petrel (barpet). He lives in coastal areas close to the sea.\n\n\n\nAnd here is a Puaiohi (puaioh). This little lady is found only on Kauai in remote forests above 4000 ft.\n\n\n\nAlas, they will probably never meet. So if your model predicts both of these birds in the same soundscape along with others, you may be able to apply logic and rule out unlikely combinations.", "votes": null }, { "id": "1692615", "postDate": "02/16/2022 06:30:20", "content": "Sorry for the noob clarification:
\nSo if the model predictions:
\nAre you suggesting, we create a meta-data of combinations and assuming A and C are never found together, we can narrow the model to A and B, thereby predicting A?
\nThere maybe outliers where A and C could be found together but we don't need a 100% accuracy π
", "rawMarkdown": "Sorry for the noob clarification:\n\nSo if the model predictions:\n\n- Bird A: 30%\n- Bird B: 25%\n- Bird C: 27%\n- Others: 18%\n\nAre you suggesting, we create a meta-data of combinations and assuming A and C are never found together, we can narrow the model to A and B, thereby predicting A?\n\nThere maybe outliers where A and C could be found together but we don't need a 100% accuracy π ", "votes": null }, { "id": "1692626", "postDate": "02/16/2022 06:42:25", "content": "Hi Sanyam. My understanding is that we are predicting at 5-second time intervals and there can be different species at different intervals (could be wrong). So suppose your example is the softmax output at t=5, and at t=10 we have Bird A: 90%, and at t=15 we have Bird B at 90%. If A & C aren't found together, but A & B are, I would conclude that C is less likely at t=5 than what the model predicted.
", "rawMarkdown": "Hi Sanyam. My understanding is that we are predicting at 5-second time intervals and there can be different species at different intervals (could be wrong). So suppose your example is the softmax output at t=5, and at t=10 we have Bird A: 90%, and at t=15 we have Bird B at 90%. If A & C aren't found together, but A & B are, I would conclude that C is less likely at t=5 than what the model predicted.", "votes": null }, { "id": "1692629", "postDate": "02/16/2022 06:43:43", "content": "Thanks so much for clarifying! π
", "rawMarkdown": "Thanks so much for clarifying! π", "votes": null }, { "id": "1692659", "postDate": "02/16/2022 06:58:55", "content": "Interesting, thanks for sharing!!!
", "rawMarkdown": "Interesting, thanks for sharing!!!", "votes": null }, { "id": "1695737", "postDate": "02/18/2022 10:39:09", "content": "impressive work @jpmiller thanks for sharing π new followerπββοΈ
", "rawMarkdown": "impressive work @jpmiller thanks for sharing π new followerπββοΈ", "votes": null } ], "comments": [ { "id": 1692615, "author_name": "init27", "author_url": "", "post_date": "02/16/2022 06:30:20", "content": "Sorry for the noob clarification:
\nSo if the model predictions:
\nAre you suggesting, we create a meta-data of combinations and assuming A and C are never found together, we can narrow the model to A and B, thereby predicting A?
\nThere maybe outliers where A and C could be found together but we don't need a 100% accuracy π
", "votes": null, "replies": [ { "id": 1692626, "author_name": "jpmiller", "author_url": "", "post_date": "02/16/2022 06:42:25", "content": "Hi Sanyam. My understanding is that we are predicting at 5-second time intervals and there can be different species at different intervals (could be wrong). So suppose your example is the softmax output at t=5, and at t=10 we have Bird A: 90%, and at t=15 we have Bird B at 90%. If A & C aren't found together, but A & B are, I would conclude that C is less likely at t=5 than what the model predicted.
", "votes": null, "replies": [] }, { "id": 1692629, "author_name": "init27", "author_url": "", "post_date": "02/16/2022 06:43:43", "content": "Thanks so much for clarifying! π
", "votes": null, "replies": [] } ] }, { "id": 1692659, "author_name": "lallucycle", "author_url": "", "post_date": "02/16/2022 06:58:55", "content": "Interesting, thanks for sharing!!!
", "votes": null, "replies": [] }, { "id": 1695737, "author_name": "arunasivapragasam", "author_url": "", "post_date": "02/18/2022 10:39:09", "content": "impressive work @jpmiller thanks for sharing π new followerπββοΈ
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1692609": "It may be possible to improve submissions by looking at the combination of birds predicted within a single soundscape. The diverse ecosystems of Hawaii lead to the separation of many species due to their preferred habitats. I lived there for 3 years and saw most every region of every island during my time there - with the exception of Ni'ihau which is exclusive to native Hawaiians. There are jungles, deserts, dry forests, swamps, lava fields, ranch land, snow-covered mountains, and of course, beaches. It makes sense that many bird species would stay within a range of ecosystems. \n\nOne example: here is a nice storm petrel (barpet). He lives in coastal areas close to the sea.\n\n\n\nAnd here is a Puaiohi (puaioh). This little lady is found only on Kauai in remote forests above 4000 ft.\n\n\n\nAlas, they will probably never meet. So if your model predicts both of these birds in the same soundscape along with others, you may be able to apply logic and rule out unlikely combinations.", "1692615": "Sorry for the noob clarification:\n\nSo if the model predictions:\n\n- Bird A: 30%\n- Bird B: 25%\n- Bird C: 27%\n- Others: 18%\n\nAre you suggesting, we create a meta-data of combinations and assuming A and C are never found together, we can narrow the model to A and B, thereby predicting A?\n\nThere maybe outliers where A and C could be found together but we don't need a 100% accuracy π ", "1692626": "Hi Sanyam. My understanding is that we are predicting at 5-second time intervals and there can be different species at different intervals (could be wrong). So suppose your example is the softmax output at t=5, and at t=10 we have Bird A: 90%, and at t=15 we have Bird B at 90%. If A & C aren't found together, but A & B are, I would conclude that C is less likely at t=5 than what the model predicted.", "1692629": "Thanks so much for clarifying! π", "1692659": "Interesting, thanks for sharing!!!", "1695737": "impressive work @jpmiller thanks for sharing π new followerπββοΈ" }, "source": "meta" }