Timeline for Can I tell my model is overfittng?
Current License: CC BY-SA 4.0
13 events
when toggle format | what | by | license | comment | |
---|---|---|---|---|---|
Jun 11, 2021 at 8:18 | comment | added | Many | This sounds reasonable. Thanks for help ! | |
Jun 11, 2021 at 8:17 | comment | added | user2974951 | Then I would test these models on the test set, for ex. the model after 100 epochs and the one after 250 epochs, and choose the one that performs best on the test set. This should give you a better estimate of how well your model performs. | |
Jun 11, 2021 at 8:13 | comment | added | Many | Yes, I have test set. | |
Jun 11, 2021 at 5:49 | comment | added | user2974951 | Do you have another set of data for testing? | |
Jun 10, 2021 at 13:03 | comment | added | Many | @user2974951 Ok then and how should I choose best model ? According to minimum in let's say epoch 100 in loss or maximum in epoch 250 according to jaccard index (mIoU) ? | |
Jun 10, 2021 at 12:48 | comment | added | user2974951 | I would say that even if your metrics are not getting worse (such as in your case, right plot), but the loss is getting worse (val loss, left plot), then there is probably overfitting going on. It might not have much of an effect on your val data at this moment, but the effect might show when you test it on another set. | |
Jun 10, 2021 at 12:35 | history | edited | Many | CC BY-SA 4.0 |
deleted 31 characters in body
|
Jun 10, 2021 at 12:32 | comment | added | Many | I would expect that loss is somehow correlated to metrics | |
Jun 10, 2021 at 12:29 | comment | added | Many | Ok so when someone is telling about overfitting, it's only according to loss function? | |
Jun 10, 2021 at 12:28 | history | edited | Many | CC BY-SA 4.0 |
added 186 characters in body
|
Jun 10, 2021 at 12:26 | comment | added | Dave | You'll pretty much always have a higher out-of-sample loss than in-sample. That should not concern you. | |
Jun 10, 2021 at 12:26 | comment | added | user2974951 | It's probably overfitting. | |
Jun 10, 2021 at 12:21 | history | asked | Many | CC BY-SA 4.0 |