Hello kagglers,
\nThis is Aditya. I knew the question might look a bit early in context of running of the competition but i wanna be clear about it very early only. I gave optizer previously, a similar competition like this, but I am not sure what to do. I haven't coded these kinds of tasks before so it's a bit challenging for me. Here, the 0.999 nb, we all knew, will not sustain or not be the number one notebook. So my question to you all is which model are you choosing or could you give some glimpses of what strategy you had for this competition. I see many models here but am not sure what to do. Also since real time data will come into picture, how can I approach to assure a good score there ? I wanna learn about this topic in more detail so created this topic.
Hello kagglers,
\nThis is Aditya. I knew the question might look a bit early in context of running of the competition but i wanna be clear about it very early only. I gave optizer previously, a similar competition like this, but I am not sure what to do. I haven't coded these kinds of tasks before so it's a bit challenging for me. Here, the 0.999 nb, we all knew, will not sustain or not be the number one notebook. So my question to you all is which model are you choosing or could you give some glimpses of what strategy you had for this competition. I see many models here but am not sure what to do. Also since real time data will come into picture, how can I approach to assure a good score there ? I wanna learn about this topic in more detail so created this topic.
Hello @adityasharma01,
\nI'm still trying to decide on the type of model. Still, I will build a few, probably Decision Tree or Boosted Decision Tree models, with a combination of Lag features; because this is a time series like the type of competition, it's essential to pass to the model information from the past.
After this, I will move into Recursive Neuronal Networks: RNN, or LSTMs.
\nI'm super curious how a DRL will work, so I will build one or more models of this type to learn more about the topic.
\nI recommend studying other competitions of this type to see the default.
Hi @adityasharma01,
\nIt is very easy to overfit financial data. Whichever model you end up trying, effective feature selection is going to be key to avoid overfitting.
\nI wrote a series of notebooks on effective feature engineering/selection: Part I, Part II, Part III.
", "rawMarkdown": "Hi @adityasharma01,\n\nIt is very easy to overfit financial data. Whichever model you end up trying, effective feature selection is going to be key to avoid overfitting.\n\nI wrote a series of notebooks on effective feature engineering/selection: [Part I](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-i), [Part II](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-ii), [Part III](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-iii).", "votes": null }, { "id": "1618755", "postDate": "12/15/2021 10:02:45", "content": "Thank you for sharing @leanboosting. Will help me a lot.
", "rawMarkdown": "Thank you for sharing @leanboosting. Will help me a lot.", "votes": null } ], "comments": [ { "id": 1607140, "author_name": "cv13j0", "author_url": "", "post_date": "12/05/2021 16:30:35", "content": "Hello @adityasharma01,
\nI'm still trying to decide on the type of model. Still, I will build a few, probably Decision Tree or Boosted Decision Tree models, with a combination of Lag features; because this is a time series like the type of competition, it's essential to pass to the model information from the past.
After this, I will move into Recursive Neuronal Networks: RNN, or LSTMs.
\nI'm super curious how a DRL will work, so I will build one or more models of this type to learn more about the topic.
\nI recommend studying other competitions of this type to see the default.
Hi @adityasharma01,
\nIt is very easy to overfit financial data. Whichever model you end up trying, effective feature selection is going to be key to avoid overfitting.
\nI wrote a series of notebooks on effective feature engineering/selection: Part I, Part II, Part III.
", "votes": null, "replies": [ { "id": 1618755, "author_name": "adityasharma01", "author_url": "", "post_date": "12/15/2021 10:02:45", "content": "Thank you for sharing @leanboosting. Will help me a lot.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1598704": "Hello kagglers, \nThis is Aditya. I knew the question might look a bit early in context of running of the competition but i wanna be clear about it very early only. I gave optizer previously, a similar competition like this, but I am not sure what to do. I haven't coded these kinds of tasks before so it's a bit challenging for me. Here, the 0.999 nb, we all knew, will not sustain or not be the number one notebook. So my question to you all is which model are you choosing or could you give some glimpses of what strategy you had for this competition. I see many models here but am not sure what to do. Also since real time data will come into picture, how can I approach to assure a good score there ? I wanna learn about this topic in more detail so created this topic.", "1607140": "Hello @adityasharma01, \nI'm still trying to decide on the type of model. Still, I will build a few, probably Decision Tree or Boosted Decision Tree models, with a combination of Lag features; because this is a time series like the type of competition, it's essential to pass to the model information from the past.\n\nAfter this, I will move into Recursive Neuronal Networks: RNN, or LSTMs.\n\nI'm super curious how a DRL will work, so I will build one or more models of this type to learn more about the topic.\nI recommend studying other competitions of this type to see the default.", "1618581": "Hi @adityasharma01,\n\nIt is very easy to overfit financial data. Whichever model you end up trying, effective feature selection is going to be key to avoid overfitting.\n\nI wrote a series of notebooks on effective feature engineering/selection: [Part I](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-i), [Part II](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-ii), [Part III](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-iii).", "1618755": "Thank you for sharing @leanboosting. Will help me a lot." }, "source": "meta" }