The primary label contains 397 types of code, but including the secondary label, there are 398 types.
\nYou can check the difference as follows.
train_df = pd.read_csv(\"../input/birdclef-2021/train_metadata.csv\")\nbirds_pri = list(train_df[\"primary_label\"].unique())\nsecondary_labels = train_df[\"secondary_labels\"].map( lambda s: s.replace(\"[\", \"\").replace(\"]\", \"\").replace(\",\", \"\").replace(\"'\", \"\").split(\" \")).values\n\nbirds_sec = []\nfor l in secondary_labels:\n for i in l:\n if i != \"\":\n birds_sec.append(i)\n\nprint(len(set(birds_pri))) # >>> 397\nprint(len(set(birds_sec))) # >>> 393\nprint(len(set(birds_sec + birds_pri))) # >>> 398\nprint(set(birds_sec + birds_pri) - set(birds_pri)) # >>> {'rocpig1'}\nMaybe the same as 'rocpig'?
", "messages": [ { "id": "1265506", "postDate": "04/07/2021 00:35:23", "content": "The primary label contains 397 types of code, but including the secondary label, there are 398 types.
\nYou can check the difference as follows.
train_df = pd.read_csv(\"../input/birdclef-2021/train_metadata.csv\")\nbirds_pri = list(train_df[\"primary_label\"].unique())\nsecondary_labels = train_df[\"secondary_labels\"].map( lambda s: s.replace(\"[\", \"\").replace(\"]\", \"\").replace(\",\", \"\").replace(\"'\", \"\").split(\" \")).values\n\nbirds_sec = []\nfor l in secondary_labels:\n for i in l:\n if i != \"\":\n birds_sec.append(i)\n\nprint(len(set(birds_pri))) # >>> 397\nprint(len(set(birds_sec))) # >>> 393\nprint(len(set(birds_sec + birds_pri))) # >>> 398\nprint(set(birds_sec + birds_pri) - set(birds_pri)) # >>> {'rocpig1'}\nMaybe the same as 'rocpig'?
", "rawMarkdown": "The primary label contains 397 types of code, but including the secondary label, there are 398 types.\nYou can check the difference as follows.\n\n``` python\ntrain_df = pd.read_csv(\"../input/birdclef-2021/train_metadata.csv\")\nbirds_pri = list(train_df[\"primary_label\"].unique())\nsecondary_labels = train_df[\"secondary_labels\"].map( lambda s: s.replace(\"[\", \"\").replace(\"]\", \"\").replace(\",\", \"\").replace(\"'\", \"\").split(\" \")).values\n\nbirds_sec = []\nfor l in secondary_labels:\n for i in l:\n if i != \"\":\n birds_sec.append(i)\n\nprint(len(set(birds_pri))) # >>> 397\nprint(len(set(birds_sec))) # >>> 393\nprint(len(set(birds_sec + birds_pri))) # >>> 398\nprint(set(birds_sec + birds_pri) - set(birds_pri)) # >>> {'rocpig1'}\n```\n\nMaybe the same as 'rocpig'?", "votes": null }, { "id": "1265798", "postDate": "04/07/2021 08:11:10", "content": "Yes and no :) The Rock Pigeon is a bit of a tricky case. eBird has two entries for this species, rocpig and rocpig1. Yet, they both refer to the same species (Columba livia). I think, it is ok to treat both codes as one class. However, you should only use primary_labels for a submission.
", "rawMarkdown": "Yes and no :) The Rock Pigeon is a bit of a tricky case. eBird has two entries for this species, *rocpig* and *rocpig1*. Yet, they both refer to the same species (*Columba livia*). I think, it is ok to treat both codes as one class. However, you should only use *primary_labels* for a submission.", "votes": null }, { "id": "1266072", "postDate": "04/07/2021 13:08:33", "content": "Thank you for the clear answer!
", "rawMarkdown": "Thank you for the clear answer!", "votes": null }, { "id": "1268904", "postDate": "04/09/2021 23:38:42", "content": "This does raise a related question. The differentiation in the Clements list between rocpig and rocpig1 seems to be wild vs. domestic. I know that no domestics are in the primary label list, so that's a non-issue.
\nHowever, if a domestic bird (such as a chicken) is heard in the background of a soundscape recording, is that a non-target sound that should be ignored? I ask because I think I hear a rooster in one or two of the Costa Rican training soundscapes, and there's no bird indicated at the relevant time stamps that I've looked at. Barking dogs suggest a rural domestic setting that would be compatible with the chicken interpretation.
\nThanks again!
", "rawMarkdown": "This does raise a related question. The differentiation in the Clements list between rocpig and rocpig1 seems to be wild vs. domestic. I know that no domestics are in the primary label list, so that's a non-issue. \n\nHowever, if a domestic bird (such as a chicken) is heard in the background of a soundscape recording, is that a non-target sound that should be ignored? I ask because I *think* I hear a rooster in one or two of the Costa Rican training soundscapes, and there's no bird indicated at the relevant time stamps that I've looked at. Barking dogs suggest a rural domestic setting that would be compatible with the chicken interpretation. \n\nThanks again!", "votes": null }, { "id": "1269217", "postDate": "04/10/2021 09:55:40", "content": "Uh, tricky one. I have to admit that I am not 100% familiar with the annotations, the only thing I know is that our annotators tried to label every bird call that they were able to identify. Yet, they might have missed some. However, in this competition, that would be a \"nocall\". All the birds that were annotated do have training data, the training data itself still contains other birds that might be likely to occur at the recording sites.
", "rawMarkdown": "Uh, tricky one. I have to admit that I am not 100% familiar with the annotations, the only thing I know is that our annotators tried to label every bird call that they were able to identify. Yet, they might have missed some. However, in this competition, that would be a \"nocall\". All the birds that were annotated do have training data, the training data itself still contains other birds that might be likely to occur at the recording sites.", "votes": null }, { "id": "1269320", "postDate": "04/10/2021 12:24:59", "content": "Thanks so much for all of your prompt responses. I have a few more annotation questions coming, but I'll try to put them together in a list that the annotators might be able to address all at once. It might be a few days.
", "rawMarkdown": "Thanks so much for all of your prompt responses. I have a few more annotation questions coming, but I'll try to put them together in a list that the annotators might be able to address all at once. It might be a few days.", "votes": null } ], "comments": [ { "id": 1265798, "author_name": "stefankahl", "author_url": "", "post_date": "04/07/2021 08:11:10", "content": "Yes and no :) The Rock Pigeon is a bit of a tricky case. eBird has two entries for this species, rocpig and rocpig1. Yet, they both refer to the same species (Columba livia). I think, it is ok to treat both codes as one class. However, you should only use primary_labels for a submission.
", "votes": null, "replies": [ { "id": 1266072, "author_name": "tattaka", "author_url": "", "post_date": "04/07/2021 13:08:33", "content": "Thank you for the clear answer!
", "votes": null, "replies": [] } ] }, { "id": 1268904, "author_name": "jmreuter", "author_url": "", "post_date": "04/09/2021 23:38:42", "content": "This does raise a related question. The differentiation in the Clements list between rocpig and rocpig1 seems to be wild vs. domestic. I know that no domestics are in the primary label list, so that's a non-issue.
\nHowever, if a domestic bird (such as a chicken) is heard in the background of a soundscape recording, is that a non-target sound that should be ignored? I ask because I think I hear a rooster in one or two of the Costa Rican training soundscapes, and there's no bird indicated at the relevant time stamps that I've looked at. Barking dogs suggest a rural domestic setting that would be compatible with the chicken interpretation.
\nThanks again!
", "votes": null, "replies": [ { "id": 1269217, "author_name": "stefankahl", "author_url": "", "post_date": "04/10/2021 09:55:40", "content": "Uh, tricky one. I have to admit that I am not 100% familiar with the annotations, the only thing I know is that our annotators tried to label every bird call that they were able to identify. Yet, they might have missed some. However, in this competition, that would be a \"nocall\". All the birds that were annotated do have training data, the training data itself still contains other birds that might be likely to occur at the recording sites.
", "votes": null, "replies": [ { "id": 1269320, "author_name": "jmreuter", "author_url": "", "post_date": "04/10/2021 12:24:59", "content": "Thanks so much for all of your prompt responses. I have a few more annotation questions coming, but I'll try to put them together in a list that the annotators might be able to address all at once. It might be a few days.
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "1265506": "The primary label contains 397 types of code, but including the secondary label, there are 398 types.\nYou can check the difference as follows.\n\n``` python\ntrain_df = pd.read_csv(\"../input/birdclef-2021/train_metadata.csv\")\nbirds_pri = list(train_df[\"primary_label\"].unique())\nsecondary_labels = train_df[\"secondary_labels\"].map( lambda s: s.replace(\"[\", \"\").replace(\"]\", \"\").replace(\",\", \"\").replace(\"'\", \"\").split(\" \")).values\n\nbirds_sec = []\nfor l in secondary_labels:\n for i in l:\n if i != \"\":\n birds_sec.append(i)\n\nprint(len(set(birds_pri))) # >>> 397\nprint(len(set(birds_sec))) # >>> 393\nprint(len(set(birds_sec + birds_pri))) # >>> 398\nprint(set(birds_sec + birds_pri) - set(birds_pri)) # >>> {'rocpig1'}\n```\n\nMaybe the same as 'rocpig'?", "1265798": "Yes and no :) The Rock Pigeon is a bit of a tricky case. eBird has two entries for this species, *rocpig* and *rocpig1*. Yet, they both refer to the same species (*Columba livia*). I think, it is ok to treat both codes as one class. However, you should only use *primary_labels* for a submission.", "1266072": "Thank you for the clear answer!", "1268904": "This does raise a related question. The differentiation in the Clements list between rocpig and rocpig1 seems to be wild vs. domestic. I know that no domestics are in the primary label list, so that's a non-issue. \n\nHowever, if a domestic bird (such as a chicken) is heard in the background of a soundscape recording, is that a non-target sound that should be ignored? I ask because I *think* I hear a rooster in one or two of the Costa Rican training soundscapes, and there's no bird indicated at the relevant time stamps that I've looked at. Barking dogs suggest a rural domestic setting that would be compatible with the chicken interpretation. \n\nThanks again!", "1269217": "Uh, tricky one. I have to admit that I am not 100% familiar with the annotations, the only thing I know is that our annotators tried to label every bird call that they were able to identify. Yet, they might have missed some. However, in this competition, that would be a \"nocall\". All the birds that were annotated do have training data, the training data itself still contains other birds that might be likely to occur at the recording sites.", "1269320": "Thanks so much for all of your prompt responses. I have a few more annotation questions coming, but I'll try to put them together in a list that the annotators might be able to address all at once. It might be a few days." }, "source": "meta" }