{ "id": 499071, "title": "Why is air temprature not a scalar?", "url": "/competitions/leap-atmospheric-physics-ai-climsim/discussion/499071", "author_name": "thoth000", "post_date": "2024-04-30T14:48:04.620000", "votes": 2, "comment_count": 4, "views": 0, "content": "

Many features such as state_t, state_q0001, state_q0002, etc. are treated as 60-dimensional vectors.

\n

I believe that time differentiation is a key theme for this task, but are these feature vectors time series data?
\nOr are they a collection of scalar information at a point in time?

", "messages": [ { "id": 2785900, "postDate": "2024-05-01T04:42:36.033Z", "content": "

The full dataset does have data for the same location over time. The competition organizers chose not to give us the timestamps, asking us to look at each column of air and make our predictions without access to data from earlier times. So, for this Kaggle competition there is no temporal dimension to use in modeling.

", "rawMarkdown": "The full dataset does have data for the same location over time. The competition organizers chose not to give us the timestamps, asking us to look at each column of air and make our predictions without access to data from earlier times. So, for this Kaggle competition there is no temporal dimension to use in modeling.", "votes": 3, "replies": [ { "id": 2785966, "postDate": "2024-05-01T05:17:23.073Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 2786104, "postDate": "2024-05-01T06:44:05.320Z", "content": "

This comment references the code for the dataset, where every 7th timestep was pulled. https://www.kaggle.com/competitions/leap-atmospheric-physics-ai-climsim/discussion/497280#2771293

\n

But since we don't have the location or the time data, I don't think it matters whether the data came from the same time or multiple times. I can't think of any way of treating the data for each column of air any differently than IID.

", "rawMarkdown": "This comment references the code for the dataset, where every 7th timestep was pulled. https://www.kaggle.com/competitions/leap-atmospheric-physics-ai-climsim/discussion/497280#2771293 \n\nBut since we don't have the location or the time data, I don't think it matters whether the data came from the same time or multiple times. I can't think of any way of treating the data for each column of air any differently than IID." } ] } ] }, { "id": 2784945, "postDate": "2024-04-30T15:06:08.237Z", "content": "

These are values for an array of heights into the atmosphere. Since they are contiguous, one option in theory would be to use a similar model that one would use on time series data, but that's up to you! :)

", "rawMarkdown": "These are values for an array of heights into the atmosphere. Since they are contiguous, one option in theory would be to use a similar model that one would use on time series data, but that's up to you! :)", "votes": 2 }, { "id": 2784910, "postDate": "2024-04-30T14:48:04.620Z", "content": "

Many features such as state_t, state_q0001, state_q0002, etc. are treated as 60-dimensional vectors.

\n

I believe that time differentiation is a key theme for this task, but are these feature vectors time series data?
\nOr are they a collection of scalar information at a point in time?

", "rawMarkdown": "Many features such as `state_t`, `state_q0001`, `state_q0002`, etc. are treated as 60-dimensional vectors.\n\nI believe that **time differentiation** is a key theme for this task, but are these feature vectors **time series data**? \nOr are they a collection of scalar information at a point in time?", "votes": 2 } ], "comments": [ { "id": 2785900, "author_name": "Ted K", "author_url": "", "post_date": "2024-05-01T04:42:36.033000", "content": "

The full dataset does have data for the same location over time. The competition organizers chose not to give us the timestamps, asking us to look at each column of air and make our predictions without access to data from earlier times. So, for this Kaggle competition there is no temporal dimension to use in modeling.

", "votes": 3, "replies": [ { "id": 2785966, "author_name": "", "author_url": "", "post_date": "2024-05-01T05:17:23.073000", "content": "", "votes": 0, "replies": [ { "id": 2786104, "author_name": "Ted K", "author_url": "", "post_date": "2024-05-01T06:44:05.320000", "content": "

This comment references the code for the dataset, where every 7th timestep was pulled. https://www.kaggle.com/competitions/leap-atmospheric-physics-ai-climsim/discussion/497280#2771293

\n

But since we don't have the location or the time data, I don't think it matters whether the data came from the same time or multiple times. I can't think of any way of treating the data for each column of air any differently than IID.

", "votes": 0, "replies": [] } ] } ] }, { "id": 2784945, "author_name": "Jekasm19", "author_url": "", "post_date": "2024-04-30T15:06:08.237000", "content": "

These are values for an array of heights into the atmosphere. Since they are contiguous, one option in theory would be to use a similar model that one would use on time series data, but that's up to you! :)

", "votes": 2, "replies": [] } ], "raw_markdown_by_id": { "2785900": "The full dataset does have data for the same location over time. The competition organizers chose not to give us the timestamps, asking us to look at each column of air and make our predictions without access to data from earlier times. So, for this Kaggle competition there is no temporal dimension to use in modeling.", "2784945": "These are values for an array of heights into the atmosphere. Since they are contiguous, one option in theory would be to use a similar model that one would use on time series data, but that's up to you! :)", "2784910": "Many features such as `state_t`, `state_q0001`, `state_q0002`, etc. are treated as 60-dimensional vectors.\n\nI believe that **time differentiation** is a key theme for this task, but are these feature vectors **time series data**? \nOr are they a collection of scalar information at a point in time?" } }