![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F16344235%2Fc1f561859ea910a3d32e8d6fcaa8b9be%2Fmotivation_image.png?generation=1758218203882359&alt=media\")
Welcome to the MABe 2025 Mouse Social Action Recognition competition!
\nThe goal of this competition is to build machine learning tools that can unite the diverse spectrum of data collected by groups studying mouse behaviors.
\n
We're a group of neuroscientists and ML researchers studying how our brains are wired for social interactions and instinctive behaviors. Several of us have made mouse action recognition systems in the past, but so far these only work in specific settings, and fail as soon as they see out-of-sample data. We hope this dataset will fuel the development of better, more robust and generalizable tools for behavioral researchers!
\nWe're very excited to host this competition, and happy to answer your questions.
\nThank you for joining us, and best of luck!
", "messages": [ { "id": 3290954, "postDate": "2025-09-18T18:03:32.507Z", "content": "Welcome to the MABe 2025 Mouse Social Action Recognition competition!
\nThe goal of this competition is to build machine learning tools that can unite the diverse spectrum of data collected by groups studying mouse behaviors.
\n
We're a group of neuroscientists and ML researchers studying how our brains are wired for social interactions and instinctive behaviors. Several of us have made mouse action recognition systems in the past, but so far these only work in specific settings, and fail as soon as they see out-of-sample data. We hope this dataset will fuel the development of better, more robust and generalizable tools for behavioral researchers!
\nWe're very excited to host this competition, and happy to answer your questions.
\nThank you for joining us, and best of luck!
", "rawMarkdown": "Welcome to the MABe 2025 Mouse Social Action Recognition competition!\n\nThe goal of this competition is to build machine learning tools that can unite the diverse spectrum of data collected by groups studying mouse behaviors.\n\n\n\nWe're a group of neuroscientists and ML researchers studying how our brains are wired for social interactions and instinctive behaviors. Several of us have made mouse action recognition systems in the past, but so far these only work in specific settings, and fail as soon as they see out-of-sample data. We hope this dataset will fuel the development of better, more robust and generalizable tools for behavioral researchers!\n\nWe're very excited to host this competition, and happy to answer your questions.\n\nThank you for joining us, and best of luck!", "votes": 27 }, { "id": 3367109, "postDate": "2025-12-08T07:56:56.173Z", "content": "Hi everyone,\nI’m still getting familiar with the dataset and competition structure before diving fully into model development. I’ve spent some time doing a deeper exploratory analysis, especially around behavior distribution, tracking completeness, and temporal patterns across the different labs. One thing that stood out is the strong class imbalance and the variation in behavior diversity between datasets.
", "rawMarkdown": "Hi everyone,\nI’m still getting familiar with the dataset and competition structure before diving fully into model development. I’ve spent some time doing a deeper exploratory analysis, especially around behavior distribution, tracking completeness, and temporal patterns across the different labs. One thing that stood out is the strong class imbalance and the variation in behavior diversity between datasets.", "replies": [ { "id": 3368308, "postDate": "2025-12-09T06:12:58.050Z", "content": "All I remember were the usage of modules that filter the data into kinematic features and agent/target ids… but i dunno how that works.
", "rawMarkdown": "All I remember were the usage of modules that filter the data into kinematic features and agent/target ids... but i dunno how that works." } ] }, { "id": 3362639, "postDate": "2025-12-05T04:05:26.120Z", "content": "Dear Kaggle Competition Organizers,\nI have a quick question regarding the test datasets used for the public and private leaderboards.\nCould you kindly confirm whether the test data (both public and private leaderboard subsets) includes any samples from laboratories whose identifiers start with \"MABe22\"?\nThank you very much for your time and support!
", "rawMarkdown": "Dear Kaggle Competition Organizers,\nI have a quick question regarding the test datasets used for the public and private leaderboards.\nCould you kindly confirm whether the test data (both public and private leaderboard subsets) includes any samples from laboratories whose identifiers start with \"MABe22\"?\nThank you very much for your time and support!", "replies": [ { "id": 3362686, "postDate": "2025-12-05T06:08:44.083Z", "content": "Nope, no MABe22 samples in either test set!
", "rawMarkdown": "Nope, no MABe22 samples in either test set!" } ] }, { "id": 3314005, "postDate": "2025-11-10T03:21:34.120Z", "content": "Hi, about the submission, I don't know how to deduce the identities of agent_id and target_id. I'm planning to use XGBClassifier for the training, and does anyone have any good approaches where I can start? Thanks.
", "rawMarkdown": "Hi, about the submission, I don't know how to deduce the identities of agent_id and target_id. I'm planning to use XGBClassifier for the training, and does anyone have any good approaches where I can start? Thanks.\n" }, { "id": 3301273, "postDate": "2025-10-13T02:18:16.553Z", "content": "Dear Organizers,
\nCould you clarify if I can use a self-trained LightGBM model (hosted publicly on GitHub) in my submission notebook?
\nTraining it directly on Kaggle hits memory limits, but I’d train it locally using only permitted data, upload the model to a public GitHub repo (meeting \"freely & publicly available\" rules), and load it via the GitHub link in the notebook for predictions.
\nDoes this comply with the competition guidelines?
\nThanks!
", "rawMarkdown": "Dear Organizers,\n\nCould you clarify if I can use a self-trained LightGBM model (hosted publicly on GitHub) in my submission notebook?\n\nTraining it directly on Kaggle hits memory limits, but I’d train it locally using only permitted data, upload the model to a public GitHub repo (meeting \"freely & publicly available\" rules), and load it via the GitHub link in the notebook for predictions.\n\nDoes this comply with the competition guidelines?\n\nThanks!", "replies": [ { "id": 3302287, "postDate": "2025-10-15T13:20:25.830Z", "content": "Yes, this is allowed.
", "rawMarkdown": "Yes, this is allowed." } ] }, { "id": 3297627, "postDate": "2025-10-03T13:38:08.363Z", "content": "Thanks ,the MABe 2025 Mouse Social Action Recognition competition!
", "rawMarkdown": "Thanks ,the MABe 2025 Mouse Social Action Recognition competition!" }, { "id": 3297511, "postDate": "2025-10-03T11:01:04Z", "content": "Hi Ann,
\nQuick question - there's a \"mouse_id\" field in train.csv. Should we expect new IDs to appear in the test set? Is it just a mouse identifier, or is it an anonymized categorical variable like they often use in trading?
\nAlso, could you clarify what \"day 1 - 00:40 (lights off)\" means in the \"conditions\" field? I understand the lights on/off part, but what is considered the starting timestep?
\nThanks!
", "rawMarkdown": "Hi Ann,\n\nQuick question - there's a \"mouse_id\" field in train.csv. Should we expect new IDs to appear in the test set? Is it just a mouse identifier, or is it an anonymized categorical variable like they often use in trading?\n\nAlso, could you clarify what \"day 1 - 00:40 (lights off)\" means in the \"conditions\" field? I understand the lights on/off part, but what is considered the starting timestep?\n\nThanks!", "replies": [ { "id": 3297703, "postDate": "2025-10-03T15:15:55.873Z", "content": "Great question! The mouse_id is just a mouse identifier (if the field is blank, it's because that information wasn't recorded during the experiment.) The mice in the test set are all different animals from those in the training set- we do this so that it's a true statistically independent test of classifier performance, since different animals can have their own idiosyncratic ways of moving that we don't want to overfit on.
\nRe conditions- the MABe22 datasets come from an experiment where animals were tracked 24 hours a day for four days, partly to study sleep/wake behavior. \"day 1 - 00:40\" means 12:40am on day 1 of the recording. Mice are nocturnal and have other circadian rhythms to their behavior, so the timestamp in this data gives you some information about what behavioral states they're likely to be in.
", "rawMarkdown": "Great question! The mouse_id is just a mouse identifier (if the field is blank, it's because that information wasn't recorded during the experiment.) The mice in the test set are all different animals from those in the training set- we do this so that it's a true statistically independent test of classifier performance, since different animals can have their own idiosyncratic ways of moving that we don't want to overfit on.\n\nRe conditions- the MABe22 datasets come from an experiment where animals were tracked 24 hours a day for four days, partly to study sleep/wake behavior. \"day 1 - 00:40\" means 12:40am on day 1 of the recording. Mice are nocturnal and have other circadian rhythms to their behavior, so the timestamp in this data gives you some information about what behavioral states they're likely to be in.", "votes": 3, "replies": [ { "id": 3297705, "postDate": "2025-10-03T15:23:36.830Z", "content": "Thanks! This saves me an extra \"mouse_id\" embedding :)
", "rawMarkdown": "Thanks! This saves me an extra \"mouse_id\" embedding :)" } ] } ] }, { "id": 3296687, "postDate": "2025-10-01T12:52:54.907Z", "content": "I am trying to submit notebook. But its been 8 hours notebook is just running, I can see it in my submission tab. What could be the issue here, I will put the image below?![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4725753%2Fdf419e59c909855a68ae444e754de7d8%2FSubmission.png?generation=1759323169875838&alt=media\")
To
\nMABe Challenge Team,
\nHi, I’m trying to implement the baseline model for social action recognition.
\nI’m having trouble with preprocessing the video frames.
\nHas anyone used the dataset augmentation method mentioned in the starter notebook?
With regards..
", "rawMarkdown": "To\nMABe Challenge Team,\nHi, I’m trying to implement the baseline model for social action recognition. \nI’m having trouble with preprocessing the video frames. \nHas anyone used the dataset augmentation method mentioned in the starter notebook?\n\nWith regards..\n", "replies": [ { "id": 3294171, "postDate": "2025-09-25T14:12:38.890Z", "content": "Thank you very much for the clarification 🙏. That helps a lot.
\nFor now, I’ll focus on the provided competition dataset to make sure I understand the pipeline correctly. Later, I might look into external datasets if it’s useful and still within the rules.
\nCould you also please confirm: when submitting, do we only need to upload our notebook that generates a submission.csv in the correct format? Or is there anything else required (such as using the starter notebook structure)?
\nThanks again for your guidance—I really appreciate the support from this community.
", "rawMarkdown": "Thank you very much for the clarification 🙏. That helps a lot.\n\nFor now, I’ll focus on the provided competition dataset to make sure I understand the pipeline correctly. Later, I might look into external datasets if it’s useful and still within the rules.\n\nCould you also please confirm: when submitting, do we only need to upload our notebook that generates a submission.csv in the correct format? Or is there anything else required (such as using the starter notebook structure)?\n\nThanks again for your guidance—I really appreciate the support from this community.\n" } ] }, { "id": 3293105, "postDate": "2025-09-23T02:25:33.980Z", "content": "Dear MABe Challenge Team,
\nI hope this message finds you well. My name is Sphesihle Qwabe, and I am participating in the MABe Challenge – Social Action Recognition in Mice.
\nI would like to clarify a question regarding the dataset. Am I required to use only the datasets provided by the competition, or am I allowed to search for and incorporate additional external datasets to improve my model? If external datasets are allowed, are there any specific guidelines or restrictions I should be aware of?
\nThank you very much for your time and guidance. I appreciate your support and look forward to your response.
", "rawMarkdown": "Dear MABe Challenge Team,\n\nI hope this message finds you well. My name is Sphesihle Qwabe, and I am participating in the MABe Challenge – Social Action Recognition in Mice.\n\nI would like to clarify a question regarding the dataset. Am I required to use only the datasets provided by the competition, or am I allowed to search for and incorporate additional external datasets to improve my model? If external datasets are allowed, are there any specific guidelines or restrictions I should be aware of?\n\nThank you very much for your time and guidance. I appreciate your support and look forward to your response.", "replies": [ { "id": 3293359, "postDate": "2025-09-23T16:49:47.763Z", "content": "As per Code Requirements under the Overview tab:
\nFreely & publicly available external data is allowed, including pre-trained models (note that we already provide formatted copies of most known external data, see below.)
So can I find my own external datasets and do everything that requires in my own notebook can I know how should I submit, can your also please tell specifically what is actually needed to do to this Competition
", "rawMarkdown": "So can I find my own external datasets and do everything that requires in my own notebook can I know how should I submit, can your also please tell specifically what is actually needed to do to this Competition " } ] } ] }, { "id": 3291888, "postDate": "2025-09-20T15:42:46.023Z", "content": "I've been doing some research for this competition, and so far it's been quite interesting. Only, I'm not sure what to do with the MABe F Beta metric notebook, is it supposed to be implemented into our own notebooks?
\nI'm talking about this notebook: https://www.kaggle.com/code/metric/mabe-f-beta I'm just confused on what to do with it.
", "rawMarkdown": "I've been doing some research for this competition, and so far it's been quite interesting. Only, I'm not sure what to do with the MABe F Beta metric notebook, is it supposed to be implemented into our own notebooks?\n\nI'm talking about this notebook: [https://www.kaggle.com/code/metric/mabe-f-beta](url) I'm just confused on what to do with it.", "replies": [ { "id": 3292518, "postDate": "2025-09-22T02:56:46.313Z", "content": "The metric notebook is to show you how your own submissions will be evaluated- you don't need to use it in your own code, but you might want to examine it to get an understanding of how scores are calculated.
", "rawMarkdown": "The metric notebook is to show you how your own submissions will be evaluated- you don't need to use it in your own code, but you might want to examine it to get an understanding of how scores are calculated." } ] }, { "id": 3294601, "postDate": "2025-09-26T11:51:15.853Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 3367109, "author_name": "Ladipo Samson", "author_url": "", "post_date": "2025-12-08T07:56:56.173000", "content": "Hi everyone,\nI’m still getting familiar with the dataset and competition structure before diving fully into model development. I’ve spent some time doing a deeper exploratory analysis, especially around behavior distribution, tracking completeness, and temporal patterns across the different labs. One thing that stood out is the strong class imbalance and the variation in behavior diversity between datasets.
", "votes": 0, "replies": [ { "id": 3368308, "author_name": "A Random Hobbyist", "author_url": "", "post_date": "2025-12-09T06:12:58.050000", "content": "All I remember were the usage of modules that filter the data into kinematic features and agent/target ids… but i dunno how that works.
", "votes": 0, "replies": [] } ] }, { "id": 3362639, "author_name": "Gnaix", "author_url": "", "post_date": "2025-12-05T04:05:26.120000", "content": "Dear Kaggle Competition Organizers,\nI have a quick question regarding the test datasets used for the public and private leaderboards.\nCould you kindly confirm whether the test data (both public and private leaderboard subsets) includes any samples from laboratories whose identifiers start with \"MABe22\"?\nThank you very much for your time and support!
", "votes": 0, "replies": [ { "id": 3362686, "author_name": "Ann Kennedy", "author_url": "", "post_date": "2025-12-05T06:08:44.083000", "content": "Nope, no MABe22 samples in either test set!
", "votes": 0, "replies": [] } ] }, { "id": 3314005, "author_name": "A Random Hobbyist", "author_url": "", "post_date": "2025-11-10T03:21:34.120000", "content": "Hi, about the submission, I don't know how to deduce the identities of agent_id and target_id. I'm planning to use XGBClassifier for the training, and does anyone have any good approaches where I can start? Thanks.
", "votes": 0, "replies": [] }, { "id": 3301273, "author_name": "Gnaix", "author_url": "", "post_date": "2025-10-13T02:18:16.553000", "content": "Dear Organizers,
\nCould you clarify if I can use a self-trained LightGBM model (hosted publicly on GitHub) in my submission notebook?
\nTraining it directly on Kaggle hits memory limits, but I’d train it locally using only permitted data, upload the model to a public GitHub repo (meeting \"freely & publicly available\" rules), and load it via the GitHub link in the notebook for predictions.
\nDoes this comply with the competition guidelines?
\nThanks!
", "votes": 0, "replies": [ { "id": 3302287, "author_name": "Ann Kennedy", "author_url": "", "post_date": "2025-10-15T13:20:25.830000", "content": "Yes, this is allowed.
", "votes": 0, "replies": [] } ] }, { "id": 3297627, "author_name": "Khushi Yadav", "author_url": "", "post_date": "2025-10-03T13:38:08.363000", "content": "Thanks ,the MABe 2025 Mouse Social Action Recognition competition!
", "votes": 0, "replies": [] }, { "id": 3297511, "author_name": "Victor Shlepov", "author_url": "", "post_date": "2025-10-03T11:01:04", "content": "Hi Ann,
\nQuick question - there's a \"mouse_id\" field in train.csv. Should we expect new IDs to appear in the test set? Is it just a mouse identifier, or is it an anonymized categorical variable like they often use in trading?
\nAlso, could you clarify what \"day 1 - 00:40 (lights off)\" means in the \"conditions\" field? I understand the lights on/off part, but what is considered the starting timestep?
\nThanks!
", "votes": 0, "replies": [ { "id": 3297703, "author_name": "Ann Kennedy", "author_url": "", "post_date": "2025-10-03T15:15:55.873000", "content": "Great question! The mouse_id is just a mouse identifier (if the field is blank, it's because that information wasn't recorded during the experiment.) The mice in the test set are all different animals from those in the training set- we do this so that it's a true statistically independent test of classifier performance, since different animals can have their own idiosyncratic ways of moving that we don't want to overfit on.
\nRe conditions- the MABe22 datasets come from an experiment where animals were tracked 24 hours a day for four days, partly to study sleep/wake behavior. \"day 1 - 00:40\" means 12:40am on day 1 of the recording. Mice are nocturnal and have other circadian rhythms to their behavior, so the timestamp in this data gives you some information about what behavioral states they're likely to be in.
", "votes": 3, "replies": [ { "id": 3297705, "author_name": "Victor Shlepov", "author_url": "", "post_date": "2025-10-03T15:23:36.830000", "content": "Thanks! This saves me an extra \"mouse_id\" embedding :)
", "votes": 0, "replies": [] } ] } ] }, { "id": 3296687, "author_name": "Pawan26 Singh", "author_url": "", "post_date": "2025-10-01T12:52:54.907000", "content": "I am trying to submit notebook. But its been 8 hours notebook is just running, I can see it in my submission tab. What could be the issue here, I will put the image below?
To
\nMABe Challenge Team,
\nHi, I’m trying to implement the baseline model for social action recognition.
\nI’m having trouble with preprocessing the video frames.
\nHas anyone used the dataset augmentation method mentioned in the starter notebook?
With regards..
", "votes": 0, "replies": [ { "id": 3294171, "author_name": "Sphesihle Qwabe", "author_url": "", "post_date": "2025-09-25T14:12:38.890000", "content": "Thank you very much for the clarification 🙏. That helps a lot.
\nFor now, I’ll focus on the provided competition dataset to make sure I understand the pipeline correctly. Later, I might look into external datasets if it’s useful and still within the rules.
\nCould you also please confirm: when submitting, do we only need to upload our notebook that generates a submission.csv in the correct format? Or is there anything else required (such as using the starter notebook structure)?
\nThanks again for your guidance—I really appreciate the support from this community.
", "votes": 0, "replies": [] } ] }, { "id": 3293105, "author_name": "Sphesihle Qwabe", "author_url": "", "post_date": "2025-09-23T02:25:33.980000", "content": "Dear MABe Challenge Team,
\nI hope this message finds you well. My name is Sphesihle Qwabe, and I am participating in the MABe Challenge – Social Action Recognition in Mice.
\nI would like to clarify a question regarding the dataset. Am I required to use only the datasets provided by the competition, or am I allowed to search for and incorporate additional external datasets to improve my model? If external datasets are allowed, are there any specific guidelines or restrictions I should be aware of?
\nThank you very much for your time and guidance. I appreciate your support and look forward to your response.
", "votes": 0, "replies": [ { "id": 3293359, "author_name": "continuingdev", "author_url": "", "post_date": "2025-09-23T16:49:47.763000", "content": "As per Code Requirements under the Overview tab:
\nFreely & publicly available external data is allowed, including pre-trained models (note that we already provide formatted copies of most known external data, see below.)
So can I find my own external datasets and do everything that requires in my own notebook can I know how should I submit, can your also please tell specifically what is actually needed to do to this Competition
", "votes": 0, "replies": [] } ] } ] }, { "id": 3291888, "author_name": "The Swimmer", "author_url": "", "post_date": "2025-09-20T15:42:46.023000", "content": "I've been doing some research for this competition, and so far it's been quite interesting. Only, I'm not sure what to do with the MABe F Beta metric notebook, is it supposed to be implemented into our own notebooks?
\nI'm talking about this notebook: https://www.kaggle.com/code/metric/mabe-f-beta I'm just confused on what to do with it.
", "votes": 0, "replies": [ { "id": 3292518, "author_name": "Ann Kennedy", "author_url": "", "post_date": "2025-09-22T02:56:46.313000", "content": "The metric notebook is to show you how your own submissions will be evaluated- you don't need to use it in your own code, but you might want to examine it to get an understanding of how scores are calculated.
", "votes": 0, "replies": [] } ] }, { "id": 3294601, "author_name": "", "author_url": "", "post_date": "2025-09-26T11:51:15.853000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3290954": "Welcome to the MABe 2025 Mouse Social Action Recognition competition!\n\nThe goal of this competition is to build machine learning tools that can unite the diverse spectrum of data collected by groups studying mouse behaviors.\n\n\n\nWe're a group of neuroscientists and ML researchers studying how our brains are wired for social interactions and instinctive behaviors. Several of us have made mouse action recognition systems in the past, but so far these only work in specific settings, and fail as soon as they see out-of-sample data. We hope this dataset will fuel the development of better, more robust and generalizable tools for behavioral researchers!\n\nWe're very excited to host this competition, and happy to answer your questions.\n\nThank you for joining us, and best of luck!", "3367109": "Hi everyone,\nI’m still getting familiar with the dataset and competition structure before diving fully into model development. I’ve spent some time doing a deeper exploratory analysis, especially around behavior distribution, tracking completeness, and temporal patterns across the different labs. One thing that stood out is the strong class imbalance and the variation in behavior diversity between datasets.", "3362639": "Dear Kaggle Competition Organizers,\nI have a quick question regarding the test datasets used for the public and private leaderboards.\nCould you kindly confirm whether the test data (both public and private leaderboard subsets) includes any samples from laboratories whose identifiers start with \"MABe22\"?\nThank you very much for your time and support!", "3314005": "Hi, about the submission, I don't know how to deduce the identities of agent_id and target_id. I'm planning to use XGBClassifier for the training, and does anyone have any good approaches where I can start? Thanks.\n", "3301273": "Dear Organizers,\n\nCould you clarify if I can use a self-trained LightGBM model (hosted publicly on GitHub) in my submission notebook?\n\nTraining it directly on Kaggle hits memory limits, but I’d train it locally using only permitted data, upload the model to a public GitHub repo (meeting \"freely & publicly available\" rules), and load it via the GitHub link in the notebook for predictions.\n\nDoes this comply with the competition guidelines?\n\nThanks!", "3297627": "Thanks ,the MABe 2025 Mouse Social Action Recognition competition!", "3297511": "Hi Ann,\n\nQuick question - there's a \"mouse_id\" field in train.csv. Should we expect new IDs to appear in the test set? Is it just a mouse identifier, or is it an anonymized categorical variable like they often use in trading?\n\nAlso, could you clarify what \"day 1 - 00:40 (lights off)\" means in the \"conditions\" field? I understand the lights on/off part, but what is considered the starting timestep?\n\nThanks!", "3296687": "I am trying to submit notebook. But its been 8 hours notebook is just running, I can see it in my submission tab. What could be the issue here, I will put the image below?", "3294117": "To\nMABe Challenge Team,\nHi, I’m trying to implement the baseline model for social action recognition. \nI’m having trouble with preprocessing the video frames. \nHas anyone used the dataset augmentation method mentioned in the starter notebook?\n\nWith regards..\n", "3293105": "Dear MABe Challenge Team,\n\nI hope this message finds you well. My name is Sphesihle Qwabe, and I am participating in the MABe Challenge – Social Action Recognition in Mice.\n\nI would like to clarify a question regarding the dataset. Am I required to use only the datasets provided by the competition, or am I allowed to search for and incorporate additional external datasets to improve my model? If external datasets are allowed, are there any specific guidelines or restrictions I should be aware of?\n\nThank you very much for your time and guidance. I appreciate your support and look forward to your response.", "3291888": "I've been doing some research for this competition, and so far it's been quite interesting. Only, I'm not sure what to do with the MABe F Beta metric notebook, is it supposed to be implemented into our own notebooks?\n\nI'm talking about this notebook: [https://www.kaggle.com/code/metric/mabe-f-beta](url) I'm just confused on what to do with it.", "3294601": "" } }