![\"Eval](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F21925130%2F3cd89952f3e779343304eb2ee4b5cad2%2FScoreMetric.png?generation=1751150274594051&alt=media\")
![\"Logloss](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F21925130%2F6f520d37099066fc296164a7ee1acc19%2FLogLoss.png?generation=1751150285894453&alt=media\")
\nGraph on Left - Competition Eval Metric
\nGraph on Right - Logloss
\n* Applied smoothing to the Graphs
\nGraph on Left - Competition Eval Metric
\nGraph on Right - Logloss
\n* Applied smoothing to the Graphs
It is a bit weird the NN model manages to achieve the best eval metric score within 5 epochs, and after than the score just fluctuates between 0.620 and 0.635. Usually you get a nice smooth curve, like seen in the logloss metric graph.
\nThe second point is that even though at the end the logloss score increases after 5 epochs, the eval metric just stays around the same, which confuses my early stop.
\nThe only normal thing about the graphs is the maximum of the eval metric graph occurs at the same number of epochs as the minimum of the logloss graph. Does anyone else have issues with unstable training or is it just me?
", "messages": [ { "id": "3235224", "postDate": "06/28/2025 22:46:21", "content": "
\nGraph on Left - Competition Eval Metric
\nGraph on Right - Logloss
\n* Applied smoothing to the Graphs
It is a bit weird the NN model manages to achieve the best eval metric score within 5 epochs, and after than the score just fluctuates between 0.620 and 0.635. Usually you get a nice smooth curve, like seen in the logloss metric graph.
\nThe second point is that even though at the end the logloss score increases after 5 epochs, the eval metric just stays around the same, which confuses my early stop.
\nThe only normal thing about the graphs is the maximum of the eval metric graph occurs at the same number of epochs as the minimum of the logloss graph. Does anyone else have issues with unstable training or is it just me?
", "rawMarkdown": "\nGraph on Left - Competition Eval Metric\nGraph on Right - Logloss \n\\* Applied smoothing to the Graphs\n\nIt is a bit weird the NN model manages to achieve the best eval metric score within 5 epochs, and after than the score just fluctuates between 0.620 and 0.635. Usually you get a nice smooth curve, like seen in the logloss metric graph.\n\nThe second point is that even though at the end the logloss score increases after 5 epochs, the eval metric just stays around the same, which confuses my early stop.\n\nThe only normal thing about the graphs is the maximum of the eval metric graph occurs at the same number of epochs as the minimum of the logloss graph. Does anyone else have issues with unstable training or is it just me?", "votes": null }, { "id": "3235262", "postDate": "06/29/2025 01:39:35", "content": "Choose scheduler and do warm-up
", "rawMarkdown": "Choose scheduler and do warm-up", "votes": null } ], "comments": [ { "id": 3235262, "author_name": "tom99763", "author_url": "", "post_date": "06/29/2025 01:39:35", "content": "Choose scheduler and do warm-up
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3235224": "\nGraph on Left - Competition Eval Metric\nGraph on Right - Logloss \n\\* Applied smoothing to the Graphs\n\nIt is a bit weird the NN model manages to achieve the best eval metric score within 5 epochs, and after than the score just fluctuates between 0.620 and 0.635. Usually you get a nice smooth curve, like seen in the logloss metric graph.\n\nThe second point is that even though at the end the logloss score increases after 5 epochs, the eval metric just stays around the same, which confuses my early stop.\n\nThe only normal thing about the graphs is the maximum of the eval metric graph occurs at the same number of epochs as the minimum of the logloss graph. Does anyone else have issues with unstable training or is it just me?", "3235262": "Choose scheduler and do warm-up" }, "source": "meta" }