I just came from the CommonLit contest and found the data size here is much larger than that and the model also takes quite long to train. So instead of 5 fold CV strategy there, I thought maybe we should do something different here: Like training 1 model using 80% train, 10% valid, 10% test? I saw the popular notebook G2Net / efficientnet_b7 / baseline [training] use 80% to train and 20% to valid but I am not sure if that's the best practice. Thank you.
", "messages": [ { "id": 1439478, "postDate": "2021-08-03T17:02:53.307Z", "content": "I've conducted almost all of my experiments using 4 Fold CV at this point and gotten pretty close to the LB. My local CV Score was 0.865 and the LB score was 0.866.
Thank you Saurav for sharing
", "rawMarkdown": "Thank you Saurav for sharing" } ] }, { "id": 1419168, "postDate": "2021-08-03T02:08:17.480Z", "content": "I just came from the CommonLit contest and found the data size here is much larger than that and the model also takes quite long to train. So instead of 5 fold CV strategy there, I thought maybe we should do something different here: Like training 1 model using 80% train, 10% valid, 10% test? I saw the popular notebook G2Net / efficientnet_b7 / baseline [training] use 80% to train and 20% to valid but I am not sure if that's the best practice. Thank you.
", "rawMarkdown": "I just came from the CommonLit contest and found the data size here is much larger than that and the model also takes quite long to train. So instead of 5 fold CV strategy there, I thought maybe we should do something different here: Like training 1 model using 80% train, 10% valid, 10% test? I saw the popular notebook [G2Net / efficientnet_b7 / baseline [training]](https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b7-baseline-training) use 80% to train and 20% to valid but I am not sure if that's the best practice. Thank you.", "votes": 1 }, { "id": 1561298, "postDate": "2021-10-27T13:13:05.380Z", "content": "Hey All,
\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.
\nThank you all,
\nRegards,
\nChris
Thank you for your reply! What do you mean by 50% discount and 7% online here? I may have missed something..
", "rawMarkdown": "Thank you for your reply! What do you mean by 50% discount and 7% online here? I may have missed something.." }, { "id": 1431000, "postDate": "2021-08-03T12:03:49.007Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1448220, "postDate": "2021-08-04T16:57:14.197Z", "content": "I tried using a 5 fold CV with EfficientNetB7 but ran out of the notebook time limit after training for a single fold. But even with this model, I was able to get my current best score of 0.866
I've conducted almost all of my experiments using 4 Fold CV at this point and gotten pretty close to the LB. My local CV Score was 0.865 and the LB score was 0.866.
Thank you Saurav for sharing
", "votes": 0, "replies": [] } ] }, { "id": 1561298, "author_name": "ChristopherZerafa", "author_url": "", "post_date": "2021-10-27T13:13:05.380000", "content": "Hey All,
\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.
\nThank you all,
\nRegards,
\nChris
Thank you for your reply! What do you mean by 50% discount and 7% online here? I may have missed something..
", "votes": 0, "replies": [] }, { "id": 1431000, "author_name": "", "author_url": "", "post_date": "2021-08-03T12:03:49.007000", "content": "", "votes": 0, "replies": [] }, { "id": 1448220, "author_name": "Saurav Maheshkar ☕️", "author_url": "", "post_date": "2021-08-04T16:57:14.197000", "content": "I tried using a 5 fold CV with EfficientNetB7 but ran out of the notebook time limit after training for a single fold. But even with this model, I was able to get my current best score of 0.866