{ "id": 285103, "title": "Question about implement rolling feature ?", "url": "/competitions/g-research-crypto-forecasting/discussion/285103", "author_name": "", "post_date": "2021-11-03T10:18:17.995322Z", "votes": 11, "comment_count": 4, "views": 0, "content": "

I want to add some rolling feature like MA, SMA to predict but the test set is only one row per asset. Any suggestion to do this ? Thanks.

", "messages": [ { "id": "1569221", "postDate": "11/03/2021 10:18:17", "content": "

I want to add some rolling feature like MA, SMA to predict but the test set is only one row per asset. Any suggestion to do this ? Thanks.

", "rawMarkdown": "I want to add some rolling feature like MA, SMA to predict but the test set is only one row per asset. Any suggestion to do this ? Thanks.", "votes": null }, { "id": "1569240", "postDate": "11/03/2021 10:25:28", "content": "

Hi @rezaroclif

\n

I suggest checking codes from Jane Street Market Prediction
\nhttps://www.kaggle.com/c/jane-street-market-prediction/code

\n

like this one
\nhttps://www.kaggle.com/a763337092/blending-tensorflow-and-pytorch?scriptVersionId=53097861&cellId=9

", "rawMarkdown": "Hi @rezaroclif \n\nI suggest checking codes from Jane Street Market Prediction\nhttps://www.kaggle.com/c/jane-street-market-prediction/code\n\n\nlike this one \nhttps://www.kaggle.com/a763337092/blending-tensorflow-and-pytorch?scriptVersionId=53097861&cellId=9", "votes": null }, { "id": "1569257", "postDate": "11/03/2021 10:36:47", "content": "

Dear @rezaroclif

\n

I am not sure if this helps but here is a very simple notebook \"Time series: A simple moving average (MA) model\", which should work for predicting just one value. Obviously it would need to be adapted for this competition API etc.

\n

All the best,
\ncarl

", "rawMarkdown": "Dear @rezaroclif \n\nI am not sure if this helps but here is a *very* simple notebook [\"Time series: A simple moving average (MA) model\"](https://www.kaggle.com/carlmcbrideellis/time-series-a-simple-moving-average-ma-model), which should work for predicting just one value. Obviously it would need to be adapted for this competition API *etc*.\n\nAll the best,\ncarl", "votes": null }, { "id": "1569670", "postDate": "11/03/2021 16:33:50", "content": "

Just save it. Just be careful because there will be times when an asset doesn't trade.

", "rawMarkdown": "Just save it. Just be careful because there will be times when an asset doesn't trade.", "votes": null }, { "id": "1569878", "postDate": "11/03/2021 19:35:22", "content": "

I think this is what we can do:

\n

As you iterate through the test set, you can build a time-series. So for the first n iterations you would not have the ability to build rolling features (and will need to rely on something else, or put a placeholder), but after those, you can build the features.

", "rawMarkdown": "I think this is what we can do:\n\nAs you iterate through the test set, you can build a time-series. So for the first n iterations you would not have the ability to build rolling features (and will need to rely on something else, or put a placeholder), but after those, you can build the features.", "votes": null } ], "comments": [ { "id": 1569240, "author_name": "faisalalsrheed", "author_url": "", "post_date": "11/03/2021 10:25:28", "content": "

Hi @rezaroclif

\n

I suggest checking codes from Jane Street Market Prediction
\nhttps://www.kaggle.com/c/jane-street-market-prediction/code

\n

like this one
\nhttps://www.kaggle.com/a763337092/blending-tensorflow-and-pytorch?scriptVersionId=53097861&cellId=9

", "votes": null, "replies": [] }, { "id": 1569257, "author_name": "carlmcbrideellis", "author_url": "", "post_date": "11/03/2021 10:36:47", "content": "

Dear @rezaroclif

\n

I am not sure if this helps but here is a very simple notebook \"Time series: A simple moving average (MA) model\", which should work for predicting just one value. Obviously it would need to be adapted for this competition API etc.

\n

All the best,
\ncarl

", "votes": null, "replies": [] }, { "id": 1569670, "author_name": "verracodeguacas", "author_url": "", "post_date": "11/03/2021 16:33:50", "content": "

Just save it. Just be careful because there will be times when an asset doesn't trade.

", "votes": null, "replies": [] }, { "id": 1569878, "author_name": "julian3833", "author_url": "", "post_date": "11/03/2021 19:35:22", "content": "

I think this is what we can do:

\n

As you iterate through the test set, you can build a time-series. So for the first n iterations you would not have the ability to build rolling features (and will need to rely on something else, or put a placeholder), but after those, you can build the features.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1569221": "I want to add some rolling feature like MA, SMA to predict but the test set is only one row per asset. Any suggestion to do this ? Thanks.", "1569240": "Hi @rezaroclif \n\nI suggest checking codes from Jane Street Market Prediction\nhttps://www.kaggle.com/c/jane-street-market-prediction/code\n\n\nlike this one \nhttps://www.kaggle.com/a763337092/blending-tensorflow-and-pytorch?scriptVersionId=53097861&cellId=9", "1569257": "Dear @rezaroclif \n\nI am not sure if this helps but here is a *very* simple notebook [\"Time series: A simple moving average (MA) model\"](https://www.kaggle.com/carlmcbrideellis/time-series-a-simple-moving-average-ma-model), which should work for predicting just one value. Obviously it would need to be adapted for this competition API *etc*.\n\nAll the best,\ncarl", "1569670": "Just save it. Just be careful because there will be times when an asset doesn't trade.", "1569878": "I think this is what we can do:\n\nAs you iterate through the test set, you can build a time-series. So for the first n iterations you would not have the ability to build rolling features (and will need to rely on something else, or put a placeholder), but after those, you can build the features." }, "source": "meta" }