{ "id": 585609, "title": "Assessing Model Convergence: Should I Continue Training?", "url": "/competitions/waveform-inversion/discussion/585609", "author_name": "", "post_date": "2025-06-21T17:55:29.602084900Z", "votes": 3, "comment_count": 1, "views": 0, "content": "

Based on the attached training and validation curves, is my model still improving, or should I stop training now? would you recommend more epochs or early stopping at this point and try other models maybe? would you share your train/val curves?

\n

\"\"

", "messages": [ { "id": "3229577", "postDate": "06/21/2025 17:55:29", "content": "

Based on the attached training and validation curves, is my model still improving, or should I stop training now? would you recommend more epochs or early stopping at this point and try other models maybe? would you share your train/val curves?

\n

\"\"

", "rawMarkdown": "Based on the attached training and validation curves, is my model still improving, or should I stop training now? would you recommend more epochs or early stopping at this point and try other models maybe? would you share your train/val curves?\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3842406%2Fbcf84a934fdb08a93012c64050f3dc42%2FScreenshot%202025-06-21%20195457.png?generation=1750528520598243&alt=media)", "votes": null }, { "id": "3229586", "postDate": "06/21/2025 18:08:07", "content": "

The evaluation loss has plateaued after around epoch 50 (You see the curve flattening with diminishing improvements). After epoch 60ish, the validation loss barely improves and in general model appears to have mostly converged. That seems as a good moment to stop because further epochs are unlikely to yield substantial improvements. Maybe experiment with data augmentation? I would suggest checking F1-score to make a final decision. Hope I helped :)

", "rawMarkdown": "The evaluation loss has plateaued after around epoch 50 (You see the curve flattening with diminishing improvements). After epoch 60ish, the validation loss barely improves and in general model appears to have mostly converged. That seems as a good moment to stop because further epochs are unlikely to yield substantial improvements. Maybe experiment with data augmentation? I would suggest checking F1-score to make a final decision. Hope I helped :)", "votes": null } ], "comments": [ { "id": 3229586, "author_name": "marcelbiezunski", "author_url": "", "post_date": "06/21/2025 18:08:07", "content": "

The evaluation loss has plateaued after around epoch 50 (You see the curve flattening with diminishing improvements). After epoch 60ish, the validation loss barely improves and in general model appears to have mostly converged. That seems as a good moment to stop because further epochs are unlikely to yield substantial improvements. Maybe experiment with data augmentation? I would suggest checking F1-score to make a final decision. Hope I helped :)

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3229577": "Based on the attached training and validation curves, is my model still improving, or should I stop training now? would you recommend more epochs or early stopping at this point and try other models maybe? would you share your train/val curves?\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3842406%2Fbcf84a934fdb08a93012c64050f3dc42%2FScreenshot%202025-06-21%20195457.png?generation=1750528520598243&alt=media)", "3229586": "The evaluation loss has plateaued after around epoch 50 (You see the curve flattening with diminishing improvements). After epoch 60ish, the validation loss barely improves and in general model appears to have mostly converged. That seems as a good moment to stop because further epochs are unlikely to yield substantial improvements. Maybe experiment with data augmentation? I would suggest checking F1-score to make a final decision. Hope I helped :)" }, "source": "meta" }