{ "id": 290806, "title": "Major Notebooks Update! ", "url": "/competitions/g-research-crypto-forecasting/discussion/290806", "author_name": "", "post_date": "2021-11-26T07:08:38.978843100Z", "votes": 34, "comment_count": 8, "views": 0, "content": "

How are you guys doing?

\n
\n

I just released a major update to my 28 notebooks for this competition!

\n
\n

Over the last week, I've been reading some of your comments and opinions regarding my notebooks collection and came up with a major update [uploading now].

\n

The update includes:

\n
    \n
  1. HUUGE cleanup and refactoring: All notebooks had been reduced greatly, they all contain much less code and hold only what is a must.
    2. \n
  2. Flag for strict loading of the data: As proposed here
    3. \n
  3. Out Of Fold competition metric calculation: All notebooks that use the CV now also calculate the competition metric Out Of Fold. I also tested the metric itself: when calculating it on the strict test - it is nearly similar to the LB score on kaggle!
    4. \n
  4. Model-specific competition metric callbacks: I've implemented callbacks for calculating the competition metric for nearly all models types I use in the notebooks. Each is custom-made and based on the numpy version introduced above.
    5. \n
  5. Feature engineering: Many notebooks now have simple feature engineering functions, these functions can differ from notebook to notebook: They are based on the overall set-up.
    6. \n
  6. Feature importance: Just like feature engineering: Many notebooks now plot the feature importance of the models they train.
    7. \n
  7. Prediction correlation over time plot: Using the out-of-fold predictions, all notebooks now plot the rolling correlation of the model's predictions to the actual target of the asset they are for.
    8. \n
  8. Faster Loading: Much had been said about the data reading speed in this competition, I've updated the data collection script to save also \"Jay\" format and made all notebooks use it be default. IT IS FASTTT!
    9. \n
\n
\n

All notebooks in the series 👇

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
CV + ModelHyperparam OptimizationTime Series ModelsFeature Engineering
Neural Network StarterMLP + AELSTMTechnical Analysis #1
LightGBM StarterLightGBMWavenetTechnical Analysis #2
Catboost StarterCatboostMultivariate-Transformer [written from scratch]Time Series Agg
XGBoost StarterXGboostN-BEATSNeutralization
Supervised AE [Janestreet 1st]Supervised AE [Janestreet 1st]DeepARQuant's Volatility Features
Transformer)Transformer⏳Target Engineering
TabNet Starter⏳Fourier Analysis
Reinforcement Learning (PPO) Starter⏳Wavelets
\n
\n

Happy Kaggling!
\nCheers^^

", "messages": [ { "id": "1596062", "postDate": "11/26/2021 07:08:38", "content": "

How are you guys doing?

\n
\n

I just released a major update to my 28 notebooks for this competition!

\n
\n

Over the last week, I've been reading some of your comments and opinions regarding my notebooks collection and came up with a major update [uploading now].

\n

The update includes:

\n
    \n
  1. HUUGE cleanup and refactoring: All notebooks had been reduced greatly, they all contain much less code and hold only what is a must.
    2. \n
  2. Flag for strict loading of the data: As proposed here
    3. \n
  3. Out Of Fold competition metric calculation: All notebooks that use the CV now also calculate the competition metric Out Of Fold. I also tested the metric itself: when calculating it on the strict test - it is nearly similar to the LB score on kaggle!
    4. \n
  4. Model-specific competition metric callbacks: I've implemented callbacks for calculating the competition metric for nearly all models types I use in the notebooks. Each is custom-made and based on the numpy version introduced above.
    5. \n
  5. Feature engineering: Many notebooks now have simple feature engineering functions, these functions can differ from notebook to notebook: They are based on the overall set-up.
    6. \n
  6. Feature importance: Just like feature engineering: Many notebooks now plot the feature importance of the models they train.
    7. \n
  7. Prediction correlation over time plot: Using the out-of-fold predictions, all notebooks now plot the rolling correlation of the model's predictions to the actual target of the asset they are for.
    8. \n
  8. Faster Loading: Much had been said about the data reading speed in this competition, I've updated the data collection script to save also \"Jay\" format and made all notebooks use it be default. IT IS FASTTT!
    9. \n
\n
\n

All notebooks in the series 👇

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
CV + ModelHyperparam OptimizationTime Series ModelsFeature Engineering
Neural Network StarterMLP + AELSTMTechnical Analysis #1
LightGBM StarterLightGBMWavenetTechnical Analysis #2
Catboost StarterCatboostMultivariate-Transformer [written from scratch]Time Series Agg
XGBoost StarterXGboostN-BEATSNeutralization
Supervised AE [Janestreet 1st]Supervised AE [Janestreet 1st]DeepARQuant's Volatility Features
Transformer)Transformer⏳Target Engineering
TabNet Starter⏳Fourier Analysis
Reinforcement Learning (PPO) Starter⏳Wavelets
\n
\n

Happy Kaggling!
\nCheers^^

", "rawMarkdown": "How are you guys doing? \n\n>I just released a major update to my 28 notebooks for this competition! \n\nOver the last week, I've been reading some of your comments and opinions regarding my notebooks collection and came up with a major update [uploading now]. \n\n**The update includes:**\n\n0. **HUUGE cleanup and refactoring**: All notebooks had been reduced greatly, they all contain much less code and hold only what is a must. \n1. Flag for **strict loading** of the data: As proposed [here](https://www.kaggle.com/julian3833/proposal-for-a-meaningful-lb-strict-lgbm)\n2. **Out Of Fold competition metric calculation**: All notebooks that use the CV now also calculate the competition metric Out Of Fold. I also tested the metric itself: when calculating it on the strict test - it is nearly similar to the LB score on kaggle! \n3. Model-specific competition metric callbacks: I've implemented callbacks for calculating the competition metric for nearly all models types I use in the notebooks. Each is custom-made and based on the numpy version introduced above. \n4. Feature engineering: Many notebooks now have simple feature engineering functions, these functions can differ from notebook to notebook: They are based on the overall set-up.\n5. Feature importance: Just like feature engineering: Many notebooks now plot the feature importance of the models they train.\n6. Prediction correlation over time plot: Using the out-of-fold predictions, all notebooks now plot the rolling correlation of the model's predictions to the actual target of the asset they are for.\n7. **Faster Loading**: Much had been said about the data reading speed in this competition, I've updated the data collection script to save also \"Jay\" format and made all notebooks use it be default. IT IS FASTTT! \n\n\n____\n\n####
All notebooks in the series 👇
\n\n| CV + Model | Hyperparam Optimization | Time Series Models | Feature Engineering |\n| --- | --- | --- | --- |\n| [Neural Network Starter](https://www.kaggle.com/yamqwe/purgedgrouptimeseries-cv-with-extra-data-nn) | [MLP + AE](https://www.kaggle.com/yamqwe/bottleneck-encoder-mlp-keras-tuner) | [LSTM](https://www.kaggle.com/yamqwe/time-series-modeling-lstm) | [Technical Analysis #1](https://www.kaggle.com/yamqwe/crypto-prediction-technical-analysis-features) |\n| [LightGBM Starter](https://www.kaggle.com/yamqwe/purgedgrouptimeseries-cv-with-extra-data-lgbm) | [LightGBM](https://www.kaggle.com/yamqwe/purged-time-series-cv-lightgbm-optuna) | [Wavenet](https://www.kaggle.com/yamqwe/time-series-modeling-wavenet) | [Technical Analysis #2](https://www.kaggle.com/yamqwe/crypto-prediction-technical-analysis-feats-2) |\n| [Catboost Starter](https://www.kaggle.com/yamqwe/purgedgrouptimeseries-cv-extra-data-catboost) | [Catboost](https://www.kaggle.com/yamqwe/purged-time-series-cv-catboost-gpu-optuna) | [Multivariate-Transformer [written from scratch]](https://www.kaggle.com/yamqwe/time-series-modeling-multivariate-transformer) | [Time Series Agg](https://www.kaggle.com/yamqwe/features-all-time-series-aggregations-ever) | \n| [XGBoost Starter](https://www.kaggle.com/yamqwe/xgb-extra-data) | [XGboost](https://www.kaggle.com/yamqwe/purged-time-series-cv-xgboost-gpu-optuna) | [N-BEATS](https://www.kaggle.com/yamqwe/crypto-forecasting-n-beats) | [Neutralization](https://www.kaggle.com/yamqwe/g-research-avoid-overfit-feature-neutralization/) |\n| [Supervised AE [Janestreet 1st]](https://www.kaggle.com/yamqwe/1st-place-of-jane-street-adapted-to-crypto) | [Supervised AE [Janestreet 1st]](https://www.kaggle.com/yamqwe/1st-place-of-jane-street-keras-tuner) | [DeepAR](https://www.kaggle.com/yamqwe/probabilistic-forecasting-deepar/) | [Quant's Volatility Features](https://www.kaggle.com/yamqwe/crypto-prediction-volatility-features) |\n| [Transformer)](https://www.kaggle.com/yamqwe/let-s-test-a-transformer) | [Transformer](https://www.kaggle.com/yamqwe/sh-tcoins-transformer-baseline) | | ⏳Target Engineering |\n| [TabNet Starter](https://www.kaggle.com/yamqwe/tabnet-cv-extra-data) | | |⏳Fourier Analysis | \n| [Reinforcement Learning (PPO) Starter](https://www.kaggle.com/yamqwe/g-research-reinforcement-learning-starter) | | | ⏳Wavelets | \n\n____ \n\n\nHappy Kaggling! \nCheers^^", "votes": null }, { "id": "1597503", "postDate": "11/27/2021 15:12:54", "content": "

I think this is something related to your notebook not the whole competition so i think you should comment it in the notebooks rather i guess

", "rawMarkdown": "I think this is something related to your notebook not the whole competition so i think you should comment it in the notebooks rather i guess", "votes": null }, { "id": "1597593", "postDate": "11/27/2021 17:15:56", "content": "

better within the time limit of submission. I once ran one forked from your notebook, timeout for scoring.

", "rawMarkdown": "better within the time limit of submission. I once ran one forked from your notebook, timeout for scoring.", "votes": null }, { "id": "1597918", "postDate": "11/28/2021 04:05:08", "content": "

Agreed he just submits using overlap not by the model

", "rawMarkdown": "Agreed he just submits using overlap not by the model", "votes": null }, { "id": "1598460", "postDate": "11/28/2021 13:43:02", "content": "

Wow Thanks for sharing!

", "rawMarkdown": "Wow Thanks for sharing!", "votes": null }, { "id": "1607301", "postDate": "12/05/2021 20:15:37", "content": "

Thanks for sharing a great collection of broad knowledge, really impressive! As an aside what do you think the out of sample winning score will be for the competition? Best. St J

", "rawMarkdown": "Thanks for sharing a great collection of broad knowledge, really impressive! As an aside what do you think the out of sample winning score will be for the competition? Best. St J", "votes": null }, { "id": "1628418", "postDate": "12/24/2021 20:39:30", "content": "

I think, you don't need to waste time on \"Target Engineering\" ,because it's much better made on catboost and most of approaches will lead to leakage
\nhttps://www.kaggle.com/yamqwe/time-series-modeling-multivariate-transformer 404

", "rawMarkdown": "I think, you don't need to waste time on \"Target Engineering\" ,because it's much better made on catboost and most of approaches will lead to leakage \nhttps://www.kaggle.com/yamqwe/time-series-modeling-multivariate-transformer 404", "votes": null }, { "id": "1630512", "postDate": "12/27/2021 11:55:27", "content": "

404 - Multivariate-Transformer [written from scratch] https://www.kaggle.com/yamqwe/time-series-modeling-multivariate-transformer
\n404 - https://www.kaggle.com/yamqwe/time-series-modeling-ts-transformer?scriptVersionId=80536396

", "rawMarkdown": "404 - Multivariate-Transformer [written from scratch] https://www.kaggle.com/yamqwe/time-series-modeling-multivariate-transformer\n404 - https://www.kaggle.com/yamqwe/time-series-modeling-ts-transformer?scriptVersionId=80536396", "votes": null }, { "id": "1781545", "postDate": "05/08/2022 16:01:34", "content": "

Hi does someone still have his notebooks ?

", "rawMarkdown": "Hi does someone still have his notebooks ?", "votes": null } ], "comments": [ { "id": 1597503, "author_name": "swaralipibose", "author_url": "", "post_date": "11/27/2021 15:12:54", "content": "

I think this is something related to your notebook not the whole competition so i think you should comment it in the notebooks rather i guess

", "votes": null, "replies": [] }, { "id": 1597593, "author_name": "dragonzhang", "author_url": "", "post_date": "11/27/2021 17:15:56", "content": "

better within the time limit of submission. I once ran one forked from your notebook, timeout for scoring.

", "votes": null, "replies": [ { "id": 1597918, "author_name": "swaralipibose", "author_url": "", "post_date": "11/28/2021 04:05:08", "content": "

Agreed he just submits using overlap not by the model

", "votes": null, "replies": [] } ] }, { "id": 1598460, "author_name": "adldotori", "author_url": "", "post_date": "11/28/2021 13:43:02", "content": "

Wow Thanks for sharing!

", "votes": null, "replies": [] }, { "id": 1607301, "author_name": "jonnydossantos", "author_url": "", "post_date": "12/05/2021 20:15:37", "content": "

Thanks for sharing a great collection of broad knowledge, really impressive! As an aside what do you think the out of sample winning score will be for the competition? Best. St J

", "votes": null, "replies": [] }, { "id": 1628418, "author_name": "vladimircape", "author_url": "", "post_date": "12/24/2021 20:39:30", "content": "

I think, you don't need to waste time on \"Target Engineering\" ,because it's much better made on catboost and most of approaches will lead to leakage
\nhttps://www.kaggle.com/yamqwe/time-series-modeling-multivariate-transformer 404

", "votes": null, "replies": [] }, { "id": 1630512, "author_name": "vladimircape", "author_url": "", "post_date": "12/27/2021 11:55:27", "content": "

404 - Multivariate-Transformer [written from scratch] https://www.kaggle.com/yamqwe/time-series-modeling-multivariate-transformer
\n404 - https://www.kaggle.com/yamqwe/time-series-modeling-ts-transformer?scriptVersionId=80536396

", "votes": null, "replies": [] }, { "id": 1781545, "author_name": "harloc971", "author_url": "", "post_date": "05/08/2022 16:01:34", "content": "

Hi does someone still have his notebooks ?

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1596062": "How are you guys doing? \n\n>I just released a major update to my 28 notebooks for this competition! \n\nOver the last week, I've been reading some of your comments and opinions regarding my notebooks collection and came up with a major update [uploading now]. \n\n**The update includes:**\n\n0. **HUUGE cleanup and refactoring**: All notebooks had been reduced greatly, they all contain much less code and hold only what is a must. \n1. Flag for **strict loading** of the data: As proposed [here](https://www.kaggle.com/julian3833/proposal-for-a-meaningful-lb-strict-lgbm)\n2. **Out Of Fold competition metric calculation**: All notebooks that use the CV now also calculate the competition metric Out Of Fold. I also tested the metric itself: when calculating it on the strict test - it is nearly similar to the LB score on kaggle! \n3. Model-specific competition metric callbacks: I've implemented callbacks for calculating the competition metric for nearly all models types I use in the notebooks. Each is custom-made and based on the numpy version introduced above. \n4. Feature engineering: Many notebooks now have simple feature engineering functions, these functions can differ from notebook to notebook: They are based on the overall set-up.\n5. Feature importance: Just like feature engineering: Many notebooks now plot the feature importance of the models they train.\n6. Prediction correlation over time plot: Using the out-of-fold predictions, all notebooks now plot the rolling correlation of the model's predictions to the actual target of the asset they are for.\n7. **Faster Loading**: Much had been said about the data reading speed in this competition, I've updated the data collection script to save also \"Jay\" format and made all notebooks use it be default. IT IS FASTTT! \n\n\n____\n\n####
All notebooks in the series 👇
\n\n| CV + Model | Hyperparam Optimization | Time Series Models | Feature Engineering |\n| --- | --- | --- | --- |\n| [Neural Network Starter](https://www.kaggle.com/yamqwe/purgedgrouptimeseries-cv-with-extra-data-nn) | [MLP + AE](https://www.kaggle.com/yamqwe/bottleneck-encoder-mlp-keras-tuner) | [LSTM](https://www.kaggle.com/yamqwe/time-series-modeling-lstm) | [Technical Analysis #1](https://www.kaggle.com/yamqwe/crypto-prediction-technical-analysis-features) |\n| [LightGBM Starter](https://www.kaggle.com/yamqwe/purgedgrouptimeseries-cv-with-extra-data-lgbm) | [LightGBM](https://www.kaggle.com/yamqwe/purged-time-series-cv-lightgbm-optuna) | [Wavenet](https://www.kaggle.com/yamqwe/time-series-modeling-wavenet) | [Technical Analysis #2](https://www.kaggle.com/yamqwe/crypto-prediction-technical-analysis-feats-2) |\n| [Catboost Starter](https://www.kaggle.com/yamqwe/purgedgrouptimeseries-cv-extra-data-catboost) | [Catboost](https://www.kaggle.com/yamqwe/purged-time-series-cv-catboost-gpu-optuna) | [Multivariate-Transformer [written from scratch]](https://www.kaggle.com/yamqwe/time-series-modeling-multivariate-transformer) | [Time Series Agg](https://www.kaggle.com/yamqwe/features-all-time-series-aggregations-ever) | \n| [XGBoost Starter](https://www.kaggle.com/yamqwe/xgb-extra-data) | [XGboost](https://www.kaggle.com/yamqwe/purged-time-series-cv-xgboost-gpu-optuna) | [N-BEATS](https://www.kaggle.com/yamqwe/crypto-forecasting-n-beats) | [Neutralization](https://www.kaggle.com/yamqwe/g-research-avoid-overfit-feature-neutralization/) |\n| [Supervised AE [Janestreet 1st]](https://www.kaggle.com/yamqwe/1st-place-of-jane-street-adapted-to-crypto) | [Supervised AE [Janestreet 1st]](https://www.kaggle.com/yamqwe/1st-place-of-jane-street-keras-tuner) | [DeepAR](https://www.kaggle.com/yamqwe/probabilistic-forecasting-deepar/) | [Quant's Volatility Features](https://www.kaggle.com/yamqwe/crypto-prediction-volatility-features) |\n| [Transformer)](https://www.kaggle.com/yamqwe/let-s-test-a-transformer) | [Transformer](https://www.kaggle.com/yamqwe/sh-tcoins-transformer-baseline) | | ⏳Target Engineering |\n| [TabNet Starter](https://www.kaggle.com/yamqwe/tabnet-cv-extra-data) | | |⏳Fourier Analysis | \n| [Reinforcement Learning (PPO) Starter](https://www.kaggle.com/yamqwe/g-research-reinforcement-learning-starter) | | | ⏳Wavelets | \n\n____ \n\n\nHappy Kaggling! \nCheers^^", "1597503": "I think this is something related to your notebook not the whole competition so i think you should comment it in the notebooks rather i guess", "1597593": "better within the time limit of submission. I once ran one forked from your notebook, timeout for scoring.", "1597918": "Agreed he just submits using overlap not by the model", "1598460": "Wow Thanks for sharing!", "1607301": "Thanks for sharing a great collection of broad knowledge, really impressive! As an aside what do you think the out of sample winning score will be for the competition? Best. St J", "1628418": "I think, you don't need to waste time on \"Target Engineering\" ,because it's much better made on catboost and most of approaches will lead to leakage \nhttps://www.kaggle.com/yamqwe/time-series-modeling-multivariate-transformer 404", "1630512": "404 - Multivariate-Transformer [written from scratch] https://www.kaggle.com/yamqwe/time-series-modeling-multivariate-transformer\n404 - https://www.kaggle.com/yamqwe/time-series-modeling-ts-transformer?scriptVersionId=80536396", "1781545": "Hi does someone still have his notebooks ?" }, "source": "meta" }