\nThe image in the repo should help clarify both questions.
\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6043604%2F957ed2d68815e3db8abef7b03304a135%2Fground_truth.png?generation=1668199923227677&alt=media\")
I was reading more about the dataset from the repo and this sentence confused me:
\n\n\nFor clicks there is only a single ground truth value for each session, which is the next aid clicked during the session (although you can still predict up to 20 aid values). The ground truth for carts and orders contains all aid values that were added to a cart and ordered respectively during the session.
\n
It brought up a couple questions:
\n12899789 has 3 clicks on different AIDs. If there's only a single ground truth value for clicks in each session, why is this the case?I was reading more about the dataset from the repo and this sentence confused me:
\n\n\nFor clicks there is only a single ground truth value for each session, which is the next aid clicked during the session (although you can still predict up to 20 aid values). The ground truth for carts and orders contains all aid values that were added to a cart and ordered respectively during the session.
\n
It brought up a couple questions:
\n12899789 has 3 clicks on different AIDs. If there's only a single ground truth value for clicks in each session, why is this the case?Hi,
\nThe image in the repo should help clarify both questions.
\n
In the image you can see a session. The lower part represents the action performed by the user.
\nIn the upper part you see what the model should predict if the session would be truncated at that step.
1.
\nSo if the session was truncated after the first interaction you should predict the item selected in the click1 action ( the next aid clicked during the session) knowing only click0.
\nIf instead it was truncated at the second step you would know only click0 and click1 and would have to predict the aid of click2.
If the next action is not a click where the session is truncated you have to predict the next aid for a click action, so if the session was truncated at the step Click2 you would have to predict the aid of Click3.
\n2.
\nAs fas as orders and carts I'm not sure I understood your question.
\nFor carts and orders you care not only about the next cart/order event but you have to predict all the future cart/orders after the truncation point.
OH that's what it means by 1 click ground truth - you're only predicting the next click, not that the test set is truncated such that there's only one click. And we're predicting all future carts and orders, regardless of truncation. That makes a lot more sense now. For some reason I didn't interpret the chart in that way at all. Thanks very much for the help!
", "rawMarkdown": "OH that's what it means by 1 click ground truth - you're only predicting the next click, not that the test set is truncated such that there's only one click. And we're predicting all future carts and orders, regardless of truncation. That makes a lot more sense now. For some reason I didn't interpret the chart in that way at all. Thanks very much for the help!", "votes": null } ], "comments": [ { "id": 2026241, "author_name": "pietromaldini1", "author_url": "", "post_date": "11/11/2022 20:58:51", "content": "Hi,
\nThe image in the repo should help clarify both questions.
\n
In the image you can see a session. The lower part represents the action performed by the user.
\nIn the upper part you see what the model should predict if the session would be truncated at that step.
1.
\nSo if the session was truncated after the first interaction you should predict the item selected in the click1 action ( the next aid clicked during the session) knowing only click0.
\nIf instead it was truncated at the second step you would know only click0 and click1 and would have to predict the aid of click2.
If the next action is not a click where the session is truncated you have to predict the next aid for a click action, so if the session was truncated at the step Click2 you would have to predict the aid of Click3.
\n2.
\nAs fas as orders and carts I'm not sure I understood your question.
\nFor carts and orders you care not only about the next cart/order event but you have to predict all the future cart/orders after the truncation point.
OH that's what it means by 1 click ground truth - you're only predicting the next click, not that the test set is truncated such that there's only one click. And we're predicting all future carts and orders, regardless of truncation. That makes a lot more sense now. For some reason I didn't interpret the chart in that way at all. Thanks very much for the help!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2026189": "I was reading more about the dataset from the [repo](https://github.com/otto-de/recsys-dataset) and this sentence confused me:\n> For clicks there is only a single ground truth value for each session, which is the next aid clicked during the session (although you can still predict up to 20 aid values). The ground truth for carts and orders contains all aid values that were added to a cart and ordered respectively during the session.\n\nIt brought up a couple questions:\n1. I checked whether any test sessions had more than 1 click event, and they do. For example, session `12899789` has 3 clicks on different AIDs. If there's only a single ground truth value for clicks in each session, why is this the case?\n2. If ground truth for carts and orders contains all aid values that were added, then does that mean these types aren't truncated? Do we need to predict values for carts/orders or do we get the same score for passing them through as-is?", "2026241": "Hi,\nThe image in the repo should help clarify both questions.\n\n\nIn the image you can see a session. The lower part represents the action performed by the user.\nIn the upper part you see what the model should predict if the session would be truncated at that step.\n\n1.\nSo if the session was truncated after the first interaction you should predict the item selected in the click1 action ( the next aid clicked during the session) knowing only click0.\nIf instead it was truncated at the second step you would know only click0 and click1 and would have to predict the aid of click2.\n\nIf the next action is not a click where the session is truncated you have to predict the next aid for a click action, so if the session was truncated at the step Click2 you would have to predict the aid of Click3.\n\n2.\nAs fas as orders and carts I'm not sure I understood your question. \nFor carts and orders you care not only about the next cart/order event but you have to predict all the future cart/orders after the truncation point.", "2026339": "OH that's what it means by 1 click ground truth - you're only predicting the next click, not that the test set is truncated such that there's only one click. And we're predicting all future carts and orders, regardless of truncation. That makes a lot more sense now. For some reason I didn't interpret the chart in that way at all. Thanks very much for the help!" }, "source": "meta" }