{ "id": 285047, "title": "[Package Summary] - Every single AutoML + Time-series package out there", "url": "/competitions/g-research-crypto-forecasting/discussion/285047", "author_name": "", "post_date": "2021-11-03T07:47:19.770000", "votes": 28, "comment_count": 0, "views": 0, "content": "

Alright! so we got a new time series competition on a dataset that is incredibly easy to overfit!

\n

Let's try to solve it with AutoML because what can possibly go wrong? 😂

\n

there has been a recent growth of time series tools such as prophet [Facebook], GluonTS [Amazon], and Forecast TCN (azure framework) [Microsoft], TBATS. they are joined by some algorithmic advancements like N-Beats, AutoArima.

\n

The claim of most tools is bold: \"solve time series\".
\nAs we all know here on kaggle, \"solving time series\" is hard when you try to solve it all the way to the private leaderboard.

\n

AutoML + Time Series summary

\n
\n

TBATS. It is often true that your daily data might exhibit
\nweekly and annual patterns. This is true for many tasks like
\ncall center volume, electricity usage, or deposit withdrawal
\nrequests.

\n
\n

TBATS is a method that accounts multiple seasonalities such as daily, weekly and annual patterns. Mainly, it uses Fourier tricks with exponential smoothing models, and scaling transformations.
\nAll automatically.

\n
\n

Prophet - Prophet is an automated curve fitting package using several linear and non-linear functions of time Prophet also takes care of seasonal effect from annual to daily granularity; that includes holiday effects.

\n
\n

GluonTS - Has a rich set of pre-built models that can be used. They particularly emphasize deep learning models. You can use their abstractions and building blocks to create your own custom time series models.

\n
\n

AtsPy - Package that attempted to incorporate \"both of each\". Built on top of Auto-Arima, TBATS, Prophet, and GluonTS.

\n
\n

AutoTS - An AutoML that uses feature search to automatically find the best models, preprocessing, and ensembling for a given dataset through genetic algorithms. It also has ensemble capabilities, a combination of metrics and cross-validation options, the ability to apply subsets and weighting, regressor generation tools, live datasets, template import and export, plotting, and a collection of data shaping parameters to round out the available feature set.

\n
\n

If you know other AutoML + time series packages that are not listed here but should, please share them in the comments and they will be added!

\n

Happy Kaggling!
\nCheers ^^

", "messages": [ { "id": 1569027, "postDate": "2021-11-03T07:47:19.770Z", "content": "

Alright! so we got a new time series competition on a dataset that is incredibly easy to overfit!

\n

Let's try to solve it with AutoML because what can possibly go wrong? 😂

\n

there has been a recent growth of time series tools such as prophet [Facebook], GluonTS [Amazon], and Forecast TCN (azure framework) [Microsoft], TBATS. they are joined by some algorithmic advancements like N-Beats, AutoArima.

\n

The claim of most tools is bold: \"solve time series\".
\nAs we all know here on kaggle, \"solving time series\" is hard when you try to solve it all the way to the private leaderboard.

\n

AutoML + Time Series summary

\n
\n

TBATS. It is often true that your daily data might exhibit
\nweekly and annual patterns. This is true for many tasks like
\ncall center volume, electricity usage, or deposit withdrawal
\nrequests.

\n
\n

TBATS is a method that accounts multiple seasonalities such as daily, weekly and annual patterns. Mainly, it uses Fourier tricks with exponential smoothing models, and scaling transformations.
\nAll automatically.

\n
\n

Prophet - Prophet is an automated curve fitting package using several linear and non-linear functions of time Prophet also takes care of seasonal effect from annual to daily granularity; that includes holiday effects.

\n
\n

GluonTS - Has a rich set of pre-built models that can be used. They particularly emphasize deep learning models. You can use their abstractions and building blocks to create your own custom time series models.

\n
\n

AtsPy - Package that attempted to incorporate \"both of each\". Built on top of Auto-Arima, TBATS, Prophet, and GluonTS.

\n
\n

AutoTS - An AutoML that uses feature search to automatically find the best models, preprocessing, and ensembling for a given dataset through genetic algorithms. It also has ensemble capabilities, a combination of metrics and cross-validation options, the ability to apply subsets and weighting, regressor generation tools, live datasets, template import and export, plotting, and a collection of data shaping parameters to round out the available feature set.

\n
\n

If you know other AutoML + time series packages that are not listed here but should, please share them in the comments and they will be added!

\n

Happy Kaggling!
\nCheers ^^

", "rawMarkdown": "Alright! so we got a new time series competition on a dataset that is incredibly easy to overfit!\n\nLet's try to solve it with AutoML because what can possibly go wrong? 😂\n\n\nthere has been a recent growth of time series tools such as [prophet](https://facebook.github.io/prophet/) [Facebook], [GluonTS](https://ts.gluon.ai/) [Amazon], and Forecast TCN (azure framework) [Microsoft], [TBATS](https://pypi.org/project/tbats/). they are joined by some algorithmic advancements like [N-Beats](https://arxiv.org/abs/1905.10437), [AutoArima](https://towardsdatascience.com/time-series-forecasting-using-auto-arima-in-python-bb83e49210cd).\n\nThe claim of most tools is bold: \"solve time series\".\nAs we all know here on kaggle, \"solving time series\" is hard when you try to solve it all the way to the private leaderboard. \n\n**AutoML + Time Series summary**\n\n____\nTBATS. It is often true that your daily data might exhibit\nweekly and annual patterns. This is true for many tasks like\ncall center volume, electricity usage, or deposit withdrawal\nrequests.\n\n____\n\n[TBATS](https://pypi.org/project/tbats/) is a method that accounts multiple seasonalities such as daily, weekly and annual patterns. Mainly, it uses Fourier tricks with exponential smoothing models, and scaling transformations. \nAll automatically. \n\n____\n\n[Prophet](https://facebook.github.io/prophet/) - Prophet is an automated curve fitting package using several linear and non-linear functions of time Prophet also takes care of seasonal effect from annual to daily granularity; that includes holiday effects. \n\n____\n\n[GluonTS](https://ts.gluon.ai/) - Has a rich set of pre-built models that can be used. They particularly emphasize deep learning models. You can use their abstractions and building blocks to create your own custom time series models. \n\n____\n\n[AtsPy](https://github.com/firmai/atspy) - Package that attempted to incorporate \"both of each\". Built on top of Auto-Arima, TBATS, Prophet, and GluonTS.\n\n____\n\n[AutoTS](https://github.com/winedarksea/AutoTS) - An AutoML that uses feature search to automatically find the best models, preprocessing, and ensembling for a given dataset through genetic algorithms. It also has ensemble capabilities, a combination of metrics and cross-validation options, the ability to apply subsets and weighting, regressor generation tools, live datasets, template import and export, plotting, and a collection of data shaping parameters to round out the available feature set.\n\n____\n\nIf you know other AutoML + time series packages that are not listed here but should, please share them in the comments and they will be added! \n\n\nHappy Kaggling! \nCheers ^^\n\n\n\n", "votes": 28 } ], "comments": [], "raw_markdown_by_id": { "1569027": "Alright! so we got a new time series competition on a dataset that is incredibly easy to overfit!\n\nLet's try to solve it with AutoML because what can possibly go wrong? 😂\n\n\nthere has been a recent growth of time series tools such as [prophet](https://facebook.github.io/prophet/) [Facebook], [GluonTS](https://ts.gluon.ai/) [Amazon], and Forecast TCN (azure framework) [Microsoft], [TBATS](https://pypi.org/project/tbats/). they are joined by some algorithmic advancements like [N-Beats](https://arxiv.org/abs/1905.10437), [AutoArima](https://towardsdatascience.com/time-series-forecasting-using-auto-arima-in-python-bb83e49210cd).\n\nThe claim of most tools is bold: \"solve time series\".\nAs we all know here on kaggle, \"solving time series\" is hard when you try to solve it all the way to the private leaderboard. \n\n**AutoML + Time Series summary**\n\n____\nTBATS. It is often true that your daily data might exhibit\nweekly and annual patterns. This is true for many tasks like\ncall center volume, electricity usage, or deposit withdrawal\nrequests.\n\n____\n\n[TBATS](https://pypi.org/project/tbats/) is a method that accounts multiple seasonalities such as daily, weekly and annual patterns. Mainly, it uses Fourier tricks with exponential smoothing models, and scaling transformations. \nAll automatically. \n\n____\n\n[Prophet](https://facebook.github.io/prophet/) - Prophet is an automated curve fitting package using several linear and non-linear functions of time Prophet also takes care of seasonal effect from annual to daily granularity; that includes holiday effects. \n\n____\n\n[GluonTS](https://ts.gluon.ai/) - Has a rich set of pre-built models that can be used. They particularly emphasize deep learning models. You can use their abstractions and building blocks to create your own custom time series models. \n\n____\n\n[AtsPy](https://github.com/firmai/atspy) - Package that attempted to incorporate \"both of each\". Built on top of Auto-Arima, TBATS, Prophet, and GluonTS.\n\n____\n\n[AutoTS](https://github.com/winedarksea/AutoTS) - An AutoML that uses feature search to automatically find the best models, preprocessing, and ensembling for a given dataset through genetic algorithms. It also has ensemble capabilities, a combination of metrics and cross-validation options, the ability to apply subsets and weighting, regressor generation tools, live datasets, template import and export, plotting, and a collection of data shaping parameters to round out the available feature set.\n\n____\n\nIf you know other AutoML + time series packages that are not listed here but should, please share them in the comments and they will be added! \n\n\nHappy Kaggling! \nCheers ^^\n\n\n\n" } }