{ "id": 442762, "title": " \"step\" Vs \"timestamp\" which column should we use?", "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/442762", "author_name": "Rohit Chaudhari", "post_date": "2023-09-24T03:35:31.489000", "votes": 1, "comment_count": 3, "views": 0, "content": "

If we are already using \"timestamp\" what more value \"step\" column adds to our model? and vice versa.
\nCan someone explain how these column add value which diferentiates them from one another?

", "messages": [ { "id": 2453449, "postDate": "2023-09-24T03:56:34.567Z", "content": "

STEP is the basis for everything (evaluation). time is generally used as an input for a feature.

\n

btw, since the onset time samples are very unbalanced, I'm starting to think about whether time is a good feature to use, which could lead to a poor upper bound on the model.

", "rawMarkdown": "STEP is the basis for everything (evaluation). time is generally used as an input for a feature.\n\nbtw, since the onset time samples are very unbalanced, I'm starting to think about whether time is a good feature to use, which could lead to a poor upper bound on the model.", "votes": 5 }, { "id": 2453433, "postDate": "2023-09-24T03:35:31.490Z", "content": "

If we are already using \"timestamp\" what more value \"step\" column adds to our model? and vice versa.
\nCan someone explain how these column add value which diferentiates them from one another?

", "rawMarkdown": "If we are already using \"timestamp\" what more value \"step\" column adds to our model? and vice versa.\nCan someone explain how these column add value which diferentiates them from one another?", "votes": 1 }, { "id": 2458532, "postDate": "2023-09-27T16:45:08.530Z", "content": "

Below are my learning so far on this:

\n
    \n
  1. Working with step is much lighter on memory in comparison to the timestamp column.
    2. \n
  2. The submission for the competition requires to have step and not the timestamp column. Reference
    3. \n
\n
,series_id,step,event,score\n,c925bb,,onset,.\n,c925bb,,wakeup,.\n,d92c9f6f8a,,onset,.\n
", "rawMarkdown": "Below are my learning so far on this:\n\n\n1. Working with `step` is much lighter on memory in comparison to the `timestamp` column. \n2. The submission for the competition requires to have `step` and not the `timestamp` column. [Reference](https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/overview/evaluation)\n\n```csv\nrow_id,series_id,step,event,score\n0,038441c925bb,100,onset,0.0\n1,038441c925bb,105,wakeup,0.0\n2,03d92c9f6f8a,80,onset,0.5\n```", "isDeleted": true, "replies": [ { "id": 2504537, "postDate": "2023-10-30T01:12:25.900Z", "content": "

I noticed that the data has duplicated step values for a given series id. I also noticed that sometimes step is different between train_series and train_events for the same timestamp. How exactly is step calculated and how is it different from timestamp?

", "rawMarkdown": "I noticed that the data has duplicated step values for a given series id. I also noticed that sometimes step is different between train_series and train_events for the same timestamp. How exactly is step calculated and how is it different from timestamp?" } ] } ], "comments": [ { "id": 2453449, "author_name": "JavaZero", "author_url": "", "post_date": "2023-09-24T03:56:34.567000", "content": "

STEP is the basis for everything (evaluation). time is generally used as an input for a feature.

\n

btw, since the onset time samples are very unbalanced, I'm starting to think about whether time is a good feature to use, which could lead to a poor upper bound on the model.

", "votes": 5, "replies": [] }, { "id": 2458532, "author_name": "Bhargav Kowshik", "author_url": "", "post_date": "2023-09-27T16:45:08.530000", "content": "

Below are my learning so far on this:

\n
    \n
  1. Working with step is much lighter on memory in comparison to the timestamp column.
    2. \n
  2. The submission for the competition requires to have step and not the timestamp column. Reference
    3. \n
\n
,series_id,step,event,score\n,c925bb,,onset,.\n,c925bb,,wakeup,.\n,d92c9f6f8a,,onset,.\n
", "votes": 0, "replies": [ { "id": 2504537, "author_name": "learningmonkey", "author_url": "", "post_date": "2023-10-30T01:12:25.900000", "content": "

I noticed that the data has duplicated step values for a given series id. I also noticed that sometimes step is different between train_series and train_events for the same timestamp. How exactly is step calculated and how is it different from timestamp?

", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "2453449": "STEP is the basis for everything (evaluation). time is generally used as an input for a feature.\n\nbtw, since the onset time samples are very unbalanced, I'm starting to think about whether time is a good feature to use, which could lead to a poor upper bound on the model.", "2453433": "If we are already using \"timestamp\" what more value \"step\" column adds to our model? and vice versa.\nCan someone explain how these column add value which diferentiates them from one another?", "2458532": "Below are my learning so far on this:\n\n\n1. Working with `step` is much lighter on memory in comparison to the `timestamp` column. \n2. The submission for the competition requires to have `step` and not the `timestamp` column. [Reference](https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/overview/evaluation)\n\n```csv\nrow_id,series_id,step,event,score\n0,038441c925bb,100,onset,0.0\n1,038441c925bb,105,wakeup,0.0\n2,03d92c9f6f8a,80,onset,0.5\n```" } }