Dear Kagglers,
\nFollowing up on discussion thread themes, here are some points of clarification on data creation and use of external data:
\nData Creation
\nWe have used a custom data formulation for this competition which draws on many different data providers, and thus will not be directly replicable from publicly available sources. We will ensure a consistent dataset formulation for the future supplemental and test data in the forecasting period as in the submission phase. In the forecasting phase, after the submission period closes, the train, supplemental and test sets will be contiguous in time, with one following the minute after the other ends - i.e. there will be no gaps or overlaps in timestamps, apart from possible missing data.
Use of External Data
\nDuring the submission phase, you are able to use external data to help with your model building and validation. However, during the forecasting phase, after the submission period closes, your models will be evaluated against future data using a consistent target formulation as per the submission phase. Therefore make sure your models are not dependent on the use of external data sources during the forecasting phase.
Thanks everyone!
", "messages": [ { "id": 1578079, "postDate": "2021-11-10T16:02:45.183Z", "content": "Dear Kagglers,
\nFollowing up on discussion thread themes, here are some points of clarification on data creation and use of external data:
\nData Creation
\nWe have used a custom data formulation for this competition which draws on many different data providers, and thus will not be directly replicable from publicly available sources. We will ensure a consistent dataset formulation for the future supplemental and test data in the forecasting period as in the submission phase. In the forecasting phase, after the submission period closes, the train, supplemental and test sets will be contiguous in time, with one following the minute after the other ends - i.e. there will be no gaps or overlaps in timestamps, apart from possible missing data.
Use of External Data
\nDuring the submission phase, you are able to use external data to help with your model building and validation. However, during the forecasting phase, after the submission period closes, your models will be evaluated against future data using a consistent target formulation as per the submission phase. Therefore make sure your models are not dependent on the use of external data sources during the forecasting phase.
Thanks everyone!
", "rawMarkdown": "Dear Kagglers,\n\nFollowing up on discussion thread themes, here are some points of clarification on data creation and use of external data:\n\n**Data Creation**\nWe have used a custom data formulation for this competition which draws on many different data providers, and thus will not be directly replicable from publicly available sources. We will ensure a consistent dataset formulation for the future supplemental and test data in the forecasting period as in the submission phase. In the forecasting phase, after the submission period closes, the train, supplemental and test sets will be contiguous in time, with one following the minute after the other ends - i.e. there will be no gaps or overlaps in timestamps, apart from possible missing data.\n \n**Use of External Data**\nDuring the submission phase, you are able to use external data to help with your model building and validation. However, during the forecasting phase, after the submission period closes, your models will be evaluated against future data using a consistent target formulation as per the submission phase. Therefore make sure your models are not dependent on the use of external data sources during the forecasting phase.\n\nThanks everyone!\n", "votes": 54 }, { "id": 1629929, "postDate": "2021-12-26T18:32:49.890Z", "content": "Major question.
\ndata which you provide include test data from Public Leaderboard
\nThe last day which used in data which is provided and data which is used in the Public Leaderboard test is \"2021-09-21\"
\nThe competitions will end and start checking on 2022-02-01.
\nQuestion: Will you use for training Notebooks that gap from \"2021-09-21\" to \"2022-02-01.\"?
\nHow will Notebooks work during checking? I hope that for training you will be using data from X data to 2022-01-31 23.59 and testing will start from 2022-02-01 00:00:00. Otherwise, all lagged features will be broken.
The gap will be filled with the supplemental_train.csv file. Per organizer, more details coming in about a week.
", "rawMarkdown": "The gap will be filled with the supplemental_train.csv file. Per organizer, more details coming in about a week.", "votes": 1 } ] }, { "id": 1653085, "postDate": "2022-01-17T08:42:15.487Z", "content": "From what I understand, our model's status will be fixed at the time of submission, and won't be retraining during the test period?
", "rawMarkdown": "From what I understand, our model's status will be fixed at the time of submission, and won't be retraining during the test period?", "votes": 1, "replies": [ { "id": 1654849, "postDate": "2022-01-18T21:40:28.617Z", "content": "The notebook you submit will be rerun in its entirety at every LB update. You submit a notebook, not a model, so if the code in the notebook retrains a model, then that is what will happen (provided it doesn't break any competition rules).
", "rawMarkdown": "The notebook you submit will be rerun in its entirety at every LB update. You submit a notebook, not a model, so if the code in the notebook retrains a model, then that is what will happen (provided it doesn't break any competition rules)." } ] }, { "id": 1578441, "postDate": "2021-11-11T03:41:16.647Z", "content": "\n\nData Creation
\n
\n We have used a custom data formulation for this competition which draws on many different data providers, and thus will not be directly replicable from publicly available sources.
Please make sure that \"custom data formulation\" has enough secrets, not some simple weighted average over public data providers, if you don't want people to reverse engineer that, replicate the data creation process on a larger dataset and gain a competitive advantage over that.
\nHowever, if you are okay with that, please let us know.
", "rawMarkdown": "> **Data Creation**\n> We have used a custom data formulation for this competition which draws on many different data providers, and thus will not be directly replicable from publicly available sources. \n\nPlease make sure that \"custom data formulation\" has enough secrets, not some simple weighted average over public data providers, if you don't want people to reverse engineer that, replicate the data creation process on a larger dataset and gain a competitive advantage over that.\n\nHowever, if you are okay with that, please let us know.\n\n", "votes": 1 }, { "id": 1578293, "postDate": "2021-11-10T20:30:23.603Z", "content": "Can you clarify a bit what do you mean by different data providers ? Isn't there a risk that we learn data providers instead of meaningfull signal ?
", "rawMarkdown": "Can you clarify a bit what do you mean by different data providers ? Isn't there a risk that we learn data providers instead of meaningfull signal ?", "votes": 1 }, { "id": 1654858, "postDate": "2022-01-18T21:57:44.730Z", "content": "Might anyone be able to provide clarification on two points for the Use of External Data part?
\nWhat is the protocol regarding utility scripts? Are these permissable? If so, do they need to be made public before the competition closes?
Above, where it says…
\n\nTherefore make sure your models are not dependent on the use of external data sources during the forecasting phase.
\n
is this just a polite warning that any external data source you use will not be updated as the competition progresses?
\nOr does it literally mean \"models that use external data sources during the forecasting phase will be breaking a competition rule\"? The Competition Rules 7.C. seems to conflict with this interpretation, saying
\n\nYou may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants.
\n
Thanks in advance.
", "rawMarkdown": "Might anyone be able to provide clarification on two points for the **Use of External Data** part?\n\n1. What is the protocol regarding utility scripts? Are these permissable? If so, do they need to be made public before the competition closes?\n\n2. Above, where it says...\n\n> Therefore make sure your models are not dependent on the use of external data sources during the forecasting phase.\n\nis this just a polite warning that any external data source you use will not be updated as the competition progresses? \nOr does it literally mean \"models that use external data sources during the forecasting phase will be breaking a competition rule\"? The [Competition Rules 7.C.](https://www.kaggle.com/c/g-research-crypto-forecasting/rules) seems to conflict with this interpretation, saying\n\n> You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants.\n\nThanks in advance." }, { "id": 1600691, "postDate": "2021-11-30T16:23:29.297Z", "content": "@andrewscherer what options do we have if we would like to call a live twitter feed during the forecasting phase? We wish to use some market sentiment signals as inputs to our model while training. And if they are unavailable during forecasting, the model wont work as well as it should.
\ncc. @narendrakunapareddy
You will not be able to use any new data during the forecasting phase.
", "rawMarkdown": "You will not be able to use any new data during the forecasting phase.", "votes": 1 }, { "id": 1610646, "postDate": "2021-12-07T11:41:12.927Z", "content": "It's mostly impossible to make a correct prediction without additional data, the theory of efficient market hypothesis, EMH not working, especially on the crypto market, which is total manipulative.
", "rawMarkdown": "It's mostly impossible to make a correct prediction without additional data, the theory of efficient market hypothesis, EMH not working, especially on the crypto market, which is total manipulative. ", "votes": 1 } ] }, { "id": 1582685, "postDate": "2021-11-15T06:51:32.980Z", "content": "How do I convert the timestamp (integer) to an actual date time?
", "rawMarkdown": "How do I convert the timestamp (integer) to an actual date time?", "replies": [ { "id": 1615230, "postDate": "2021-12-12T03:33:44.827Z", "content": "The timestamp is the number of seconds since some date. You can convert the timestamp string value to a datetime using something like this python snippet.
\n df['timestamp2'] = pd.to_numeric(df['timestamp'])
\n df['DateVal'] = pd.to_datetime(df['timestamp2'], unit='s')
Major question.
\ndata which you provide include test data from Public Leaderboard
\nThe last day which used in data which is provided and data which is used in the Public Leaderboard test is \"2021-09-21\"
\nThe competitions will end and start checking on 2022-02-01.
\nQuestion: Will you use for training Notebooks that gap from \"2021-09-21\" to \"2022-02-01.\"?
\nHow will Notebooks work during checking? I hope that for training you will be using data from X data to 2022-01-31 23.59 and testing will start from 2022-02-01 00:00:00. Otherwise, all lagged features will be broken.
The gap will be filled with the supplemental_train.csv file. Per organizer, more details coming in about a week.
", "votes": 1, "replies": [] } ] }, { "id": 1653085, "author_name": "Estchen", "author_url": "", "post_date": "2022-01-17T08:42:15.487000", "content": "From what I understand, our model's status will be fixed at the time of submission, and won't be retraining during the test period?
", "votes": 1, "replies": [ { "id": 1654849, "author_name": "jagofc", "author_url": "", "post_date": "2022-01-18T21:40:28.617000", "content": "The notebook you submit will be rerun in its entirety at every LB update. You submit a notebook, not a model, so if the code in the notebook retrains a model, then that is what will happen (provided it doesn't break any competition rules).
", "votes": 0, "replies": [] } ] }, { "id": 1578441, "author_name": "Wenrui Kong", "author_url": "", "post_date": "2021-11-11T03:41:16.647000", "content": "\n\nData Creation
\n
\n We have used a custom data formulation for this competition which draws on many different data providers, and thus will not be directly replicable from publicly available sources.
Please make sure that \"custom data formulation\" has enough secrets, not some simple weighted average over public data providers, if you don't want people to reverse engineer that, replicate the data creation process on a larger dataset and gain a competitive advantage over that.
\nHowever, if you are okay with that, please let us know.
", "votes": 1, "replies": [] }, { "id": 1578293, "author_name": "Lucas Morin", "author_url": "", "post_date": "2021-11-10T20:30:23.603000", "content": "Can you clarify a bit what do you mean by different data providers ? Isn't there a risk that we learn data providers instead of meaningfull signal ?
", "votes": 1, "replies": [] }, { "id": 1654858, "author_name": "jagofc", "author_url": "", "post_date": "2022-01-18T21:57:44.730000", "content": "Might anyone be able to provide clarification on two points for the Use of External Data part?
\nWhat is the protocol regarding utility scripts? Are these permissable? If so, do they need to be made public before the competition closes?
Above, where it says…
\n\nTherefore make sure your models are not dependent on the use of external data sources during the forecasting phase.
\n
is this just a polite warning that any external data source you use will not be updated as the competition progresses?
\nOr does it literally mean \"models that use external data sources during the forecasting phase will be breaking a competition rule\"? The Competition Rules 7.C. seems to conflict with this interpretation, saying
\n\nYou may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants.
\n
Thanks in advance.
", "votes": 0, "replies": [] }, { "id": 1600691, "author_name": "Nitin Pasumarthy", "author_url": "", "post_date": "2021-11-30T16:23:29.297000", "content": "@andrewscherer what options do we have if we would like to call a live twitter feed during the forecasting phase? We wish to use some market sentiment signals as inputs to our model while training. And if they are unavailable during forecasting, the model wont work as well as it should.
\ncc. @narendrakunapareddy
You will not be able to use any new data during the forecasting phase.
", "votes": 1, "replies": [] }, { "id": 1610646, "author_name": "Vladimir Shebunyaev", "author_url": "", "post_date": "2021-12-07T11:41:12.927000", "content": "It's mostly impossible to make a correct prediction without additional data, the theory of efficient market hypothesis, EMH not working, especially on the crypto market, which is total manipulative.
", "votes": 1, "replies": [] } ] }, { "id": 1582685, "author_name": "Ruairc", "author_url": "", "post_date": "2021-11-15T06:51:32.980000", "content": "How do I convert the timestamp (integer) to an actual date time?
", "votes": 0, "replies": [ { "id": 1615230, "author_name": "MarkFowler", "author_url": "", "post_date": "2021-12-12T03:33:44.827000", "content": "The timestamp is the number of seconds since some date. You can convert the timestamp string value to a datetime using something like this python snippet.
\n df['timestamp2'] = pd.to_numeric(df['timestamp'])
\n df['DateVal'] = pd.to_datetime(df['timestamp2'], unit='s')