Hi, I am wondering if there is any possible way to combine the tags and other features to achieve a better result, would you mind showing any possible method or some hint? Thank you!
", "messages": [ { "id": "1613485", "postDate": "12/10/2021 03:04:47", "content": "Hi, I am wondering if there is any possible way to combine the tags and other features to achieve a better result, would you mind showing any possible method or some hint? Thank you!
", "rawMarkdown": "Hi, I am wondering if there is any possible way to combine the tags and other features to achieve a better result, would you mind showing any possible method or some hint? Thank you!", "votes": null }, { "id": "1613598", "postDate": "12/10/2021 06:01:48", "content": "Based on my experiment, tags could be incorporated into the ResNet features (images side) or used alone as a target since tags are attached to images in the testing set.
\nHowever, I tested incorporating tags to the ResNet features and used it alone, the results are very similar to using ResNet features alone as the target. I posted this finding in another discussion, James' reply made a lot of sense to me. Since the ResNet features already involved the characteristics of the objects in the images, thus adding another set of tags that describe the items in the image probably won't do much to the model I use.
\nAbove is my finding with the baseline model, looking forward to other people's ideas on tags!
", "rawMarkdown": "Based on my experiment, tags could be incorporated into the ResNet features (images side) or used alone as a target since tags are attached to images in the testing set.\n\nHowever, I tested incorporating tags to the ResNet features and used it alone, the results are very similar to using ResNet features **alone** as the target. I posted this finding in another discussion, James' reply made a lot of sense to me. Since the ResNet features already involved the characteristics of the objects in the images, thus adding another set of tags that describe the items in the image probably won't do much to the model I use. \n\nAbove is my finding with the baseline model, looking forward to other people's ideas on tags!", "votes": null }, { "id": "1614345", "postDate": "12/11/2021 00:37:24", "content": "When incorporating tags as features I personally didn't see any improvement.
\nI may have done something weird as well though because sometimes I saw a very slight (likely by chance) decrease in performance.
As of now I'm just not using tags at all to simplify the code.
", "rawMarkdown": "When incorporating tags as features I personally didn't see any improvement. \nI may have done something weird as well though because sometimes I saw a very slight (likely by chance) decrease in performance. \n\nAs of now I'm just not using tags at all to simplify the code.", "votes": null }, { "id": "1614875", "postDate": "12/11/2021 14:31:43", "content": "wondering too lol
", "rawMarkdown": "wondering too lol", "votes": null } ], "comments": [ { "id": 1613598, "author_name": "carolineming", "author_url": "", "post_date": "12/10/2021 06:01:48", "content": "Based on my experiment, tags could be incorporated into the ResNet features (images side) or used alone as a target since tags are attached to images in the testing set.
\nHowever, I tested incorporating tags to the ResNet features and used it alone, the results are very similar to using ResNet features alone as the target. I posted this finding in another discussion, James' reply made a lot of sense to me. Since the ResNet features already involved the characteristics of the objects in the images, thus adding another set of tags that describe the items in the image probably won't do much to the model I use.
\nAbove is my finding with the baseline model, looking forward to other people's ideas on tags!
", "votes": null, "replies": [ { "id": 1614345, "author_name": "danishvirani", "author_url": "", "post_date": "12/11/2021 00:37:24", "content": "When incorporating tags as features I personally didn't see any improvement.
\nI may have done something weird as well though because sometimes I saw a very slight (likely by chance) decrease in performance.
As of now I'm just not using tags at all to simplify the code.
", "votes": null, "replies": [] } ] }, { "id": 1614875, "author_name": "siyichen12138", "author_url": "", "post_date": "12/11/2021 14:31:43", "content": "wondering too lol
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1613485": "Hi, I am wondering if there is any possible way to combine the tags and other features to achieve a better result, would you mind showing any possible method or some hint? Thank you!", "1613598": "Based on my experiment, tags could be incorporated into the ResNet features (images side) or used alone as a target since tags are attached to images in the testing set.\n\nHowever, I tested incorporating tags to the ResNet features and used it alone, the results are very similar to using ResNet features **alone** as the target. I posted this finding in another discussion, James' reply made a lot of sense to me. Since the ResNet features already involved the characteristics of the objects in the images, thus adding another set of tags that describe the items in the image probably won't do much to the model I use. \n\nAbove is my finding with the baseline model, looking forward to other people's ideas on tags!", "1614345": "When incorporating tags as features I personally didn't see any improvement. \nI may have done something weird as well though because sometimes I saw a very slight (likely by chance) decrease in performance. \n\nAs of now I'm just not using tags at all to simplify the code.", "1614875": "wondering too lol" }, "source": "meta" }