![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2125251%2Fa62849f56cb0bc43eb4cb562c819ef0a%2FScreenshot%202023-05-20%20at%208.55.05%20PM.png?generation=1684587502315436&alt=media\")
I'm wondering if it's possible for a dot plot to admit zero counts in the context of this competition? if so, what should be the convention for predicting the data-series? For example in the plot below,
\n
Do we include 0, 1, 2 and 4 in the predicted x series (the above image is from google search, not training data of this competition)? Thanks!
", "messages": [ { "id": 2266902, "postDate": "2023-05-20T13:02:49.013Z", "content": "I'm wondering if it's possible for a dot plot to admit zero counts in the context of this competition? if so, what should be the convention for predicting the data-series? For example in the plot below,
\n
Do we include 0, 1, 2 and 4 in the predicted x series (the above image is from google search, not training data of this competition)? Thanks!
", "rawMarkdown": "I'm wondering if it's possible for a dot plot to admit zero counts in the context of this competition? if so, what should be the convention for predicting the data-series? For example in the plot below,\n\n\n\nDo we include 0, 1, 2 and 4 in the predicted x series (the above image is from google search, not training data of this competition)? Thanks!", "votes": 5 }, { "id": 2267052, "postDate": "2023-05-20T14:48:17.523Z", "content": "After looking through the data, there are some samples that contain x labels with zero dots. The annotations of such samples confirms that one must predict zeros for those x labels. I published a rule-based solution that handles this issue and it performs well for out-of-distribution samples.
", "rawMarkdown": "After looking through the data, there are some samples that contain x labels with zero dots. The annotations of such samples confirms that one must predict zeros for those x labels. I published a rule-based [solution](https://www.kaggle.com/code/awqatak/dotsolver-a-rule-based-solution-for-dot-plot) that handles this issue and it performs well for out-of-distribution samples.", "votes": 4, "replies": [ { "id": 2267053, "postDate": "2023-05-20T14:49:01.923Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2267055, "postDate": "2023-05-20T14:50:00.867Z", "content": "I think this might be the best way to handle it unfortunately. Good work on finding this!
", "rawMarkdown": "I think this might be the best way to handle it unfortunately. Good work on finding this!", "votes": 3 }, { "id": 2267097, "postDate": "2023-05-20T15:33:43.713Z", "content": "@awqatak Nice work on rule based algorithm for dot plots & many thanks for your comment! Indeed, there are some dot plots having zero counts which I missed before e.g. f6ed53eb622e
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2125251%2Fae0b0a2ddae1a21e9ba25dfcac5d1cfe%2FScreenshot%202023-05-20%20at%2011.23.04%20PM.png?generation=1684596263736777&alt=media\")
The count corresponding to 530 is zero. But, I think there is still some ambiguity, specifically when x axis is numerical. In the example above, why not include points like (520.5, 0), (521.5, 0), (522.5, 0) ….?
", "rawMarkdown": "@awqatak Nice work on rule based algorithm for dot plots & many thanks for your comment! Indeed, there are some dot plots having zero counts which I missed before e.g. `f6ed53eb622e`\n\n\nThe count corresponding to 530 is zero. But, I think there is still some ambiguity, specifically when x axis is numerical. In the example above, why not include points like (520.5, 0), (521.5, 0), (522.5, 0) ....?", "votes": 1, "replies": [ { "id": 2267188, "postDate": "2023-05-20T16:53:07.867Z", "content": "For the above dot plot, you will find that the true y-series is 6;8;2;9;10;7;8;4;8;0 which matches the prediction of DotSolver. I think there is no need to output zeros for the midpoints since this will give high redundancy.
However, your assumption of ambiguity is correct. If we assume that the above example has dots in 7 midpoints out of the 9 midpoints, the output of the y-series will have a length of 16 instead of the actual length of 19. I will try to fix this later. Thanks for pointing this out.
", "rawMarkdown": "For the above dot plot, you will find that the true y-series is `6;8;2;9;10;7;8;4;8;0` which matches the prediction of DotSolver. I think there is no need to output zeros for the midpoints since this will give high redundancy. \n\nHowever, your assumption of ambiguity is correct. If we assume that the above example has dots in 7 midpoints out of the 9 midpoints, the output of the y-series will have a length of 16 instead of the actual length of 19. I will try to fix this later. Thanks for pointing this out.", "votes": 1 } ] } ] }, { "id": 2267046, "postDate": "2023-05-20T14:45:10.230Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 2267052, "author_name": "awqatak", "author_url": "", "post_date": "2023-05-20T14:48:17.523000", "content": "After looking through the data, there are some samples that contain x labels with zero dots. The annotations of such samples confirms that one must predict zeros for those x labels. I published a rule-based solution that handles this issue and it performs well for out-of-distribution samples.
", "votes": 4, "replies": [ { "id": 2267053, "author_name": "", "author_url": "", "post_date": "2023-05-20T14:49:01.923000", "content": "", "votes": 0, "replies": [] }, { "id": 2267055, "author_name": "Cody_Null", "author_url": "", "post_date": "2023-05-20T14:50:00.867000", "content": "I think this might be the best way to handle it unfortunately. Good work on finding this!
", "votes": 3, "replies": [] }, { "id": 2267097, "author_name": "Raja Biswas", "author_url": "", "post_date": "2023-05-20T15:33:43.713000", "content": "@awqatak Nice work on rule based algorithm for dot plots & many thanks for your comment! Indeed, there are some dot plots having zero counts which I missed before e.g. f6ed53eb622e
\n
The count corresponding to 530 is zero. But, I think there is still some ambiguity, specifically when x axis is numerical. In the example above, why not include points like (520.5, 0), (521.5, 0), (522.5, 0) ….?
", "votes": 1, "replies": [ { "id": 2267188, "author_name": "awqatak", "author_url": "", "post_date": "2023-05-20T16:53:07.867000", "content": "For the above dot plot, you will find that the true y-series is 6;8;2;9;10;7;8;4;8;0 which matches the prediction of DotSolver. I think there is no need to output zeros for the midpoints since this will give high redundancy.
However, your assumption of ambiguity is correct. If we assume that the above example has dots in 7 midpoints out of the 9 midpoints, the output of the y-series will have a length of 16 instead of the actual length of 19. I will try to fix this later. Thanks for pointing this out.
", "votes": 1, "replies": [] } ] } ] }, { "id": 2267046, "author_name": "", "author_url": "", "post_date": "2023-05-20T14:45:10.230000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2266902": "I'm wondering if it's possible for a dot plot to admit zero counts in the context of this competition? if so, what should be the convention for predicting the data-series? For example in the plot below,\n\n\n\nDo we include 0, 1, 2 and 4 in the predicted x series (the above image is from google search, not training data of this competition)? Thanks!", "2267052": "After looking through the data, there are some samples that contain x labels with zero dots. The annotations of such samples confirms that one must predict zeros for those x labels. I published a rule-based [solution](https://www.kaggle.com/code/awqatak/dotsolver-a-rule-based-solution-for-dot-plot) that handles this issue and it performs well for out-of-distribution samples.", "2267046": "" } }