Open ijyliu opened 2 months ago
here's permutation improtance for logistic regression on all features. @CalBlitz do you wanna run your plot script on this
im pretty solid in us saying vgg is most important and thus deserving to be run on its own, given it's low dimension compared to vit
That's what I've been saying. Great we have the feature importance thing. Not having ResNet feels like such a bummer.
I have ResNet ready if you wanna use only that with other classifiers.
We still have overfit. Can you make a version of the data with VGG + 2 simple features
On Sun, Apr 14, 2024, 3:41 PM Gaurav Sharan Srivastava < @.***> wrote:
I have ResNet ready if you wanna use only that with other classifiers.
— Reply to this email directly, view it on GitHub https://github.com/ijyliu/computer-vision-project/issues/70#issuecomment-2054207637, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQCGE4I37WH2BLA6R7VJULDY5MA2JAVCNFSM6AAAAABF5XEG66VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANJUGIYDONRTG4 . You are receiving this because you were assigned.Message ID: @.***>
The two most important simple features
On Sun, Apr 14, 2024, 3:56 PM Isaac Liu @.***> wrote:
We still have overfit. Can you make a version of the data with VGG + 2 simple features
On Sun, Apr 14, 2024, 3:41 PM Gaurav Sharan Srivastava < @.***> wrote:
I have ResNet ready if you wanna use only that with other classifiers.
— Reply to this email directly, view it on GitHub https://github.com/ijyliu/computer-vision-project/issues/70#issuecomment-2054207637, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQCGE4I37WH2BLA6R7VJULDY5MA2JAVCNFSM6AAAAABF5XEG66VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANJUGIYDONRTG4 . You are receiving this because you were assigned.Message ID: @.***>
Er wait hang on let me check what's allowed
On Sun, Apr 14, 2024, 3:56 PM Isaac Liu @.***> wrote:
The two most important simple features
On Sun, Apr 14, 2024, 3:56 PM Isaac Liu @.***> wrote:
We still have overfit. Can you make a version of the data with VGG + 2 simple features
On Sun, Apr 14, 2024, 3:41 PM Gaurav Sharan Srivastava < @.***> wrote:
I have ResNet ready if you wanna use only that with other classifiers.
— Reply to this email directly, view it on GitHub https://github.com/ijyliu/computer-vision-project/issues/70#issuecomment-2054207637, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQCGE4I37WH2BLA6R7VJULDY5MA2JAVCNFSM6AAAAABF5XEG66VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANJUGIYDONRTG4 . You are receiving this because you were assigned.Message ID: @.***>
yeah vgg + two most important simple features.
On Sun, Apr 14, 2024 at 3:58 PM Isaac Liu @.***> wrote:
Er wait hang on let me check what's allowed
On Sun, Apr 14, 2024, 3:56 PM Isaac Liu @.***> wrote:
The two most important simple features
On Sun, Apr 14, 2024, 3:56 PM Isaac Liu @.***> wrote:
We still have overfit. Can you make a version of the data with VGG + 2 simple features
On Sun, Apr 14, 2024, 3:41 PM Gaurav Sharan Srivastava < @.***> wrote:
I have ResNet ready if you wanna use only that with other classifiers.
— Reply to this email directly, view it on GitHub https://github.com/ijyliu/computer-vision-project/issues/70#issuecomment-2054207637, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQCGE4I37WH2BLA6R7VJULDY5MA2JAVCNFSM6AAAAABF5XEG66VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANJUGIYDONRTG4 . You are receiving this because you were assigned.Message ID: @.***>
i would say do pca but actually it seems like the pca seems to damage test accuracy pretty consistently
On Sun, Apr 14, 2024 at 4:03 PM Isaac Liu @.***> wrote:
yeah vgg + two most important simple features.
On Sun, Apr 14, 2024 at 3:58 PM Isaac Liu @.***> wrote:
Er wait hang on let me check what's allowed
On Sun, Apr 14, 2024, 3:56 PM Isaac Liu @.***> wrote:
The two most important simple features
On Sun, Apr 14, 2024, 3:56 PM Isaac Liu @.***> wrote:
We still have overfit. Can you make a version of the data with VGG + 2 simple features
On Sun, Apr 14, 2024, 3:41 PM Gaurav Sharan Srivastava < @.***> wrote:
I have ResNet ready if you wanna use only that with other classifiers.
— Reply to this email directly, view it on GitHub https://github.com/ijyliu/computer-vision-project/issues/70#issuecomment-2054207637, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQCGE4I37WH2BLA6R7VJULDY5MA2JAVCNFSM6AAAAABF5XEG66VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANJUGIYDONRTG4 . You are receiving this because you were assigned.Message ID: @.***>
Unless the two most important includes HOG as one of them. Lol. Basically not worth it then
So I guess make the importance plot on that logistic reg data i sent and if hog is least important we can keep trying
On Sun, Apr 14, 2024, 4:04 PM Isaac Liu @.***> wrote:
i would say do pca but actually it seems like the pca seems to damage test accuracy pretty consistently
On Sun, Apr 14, 2024 at 4:03 PM Isaac Liu @.***> wrote:
yeah vgg + two most important simple features.
On Sun, Apr 14, 2024 at 3:58 PM Isaac Liu @.***> wrote:
Er wait hang on let me check what's allowed
On Sun, Apr 14, 2024, 3:56 PM Isaac Liu @.***> wrote:
The two most important simple features
On Sun, Apr 14, 2024, 3:56 PM Isaac Liu @.***> wrote:
We still have overfit. Can you make a version of the data with VGG + 2 simple features
On Sun, Apr 14, 2024, 3:41 PM Gaurav Sharan Srivastava < @.***> wrote:
I have ResNet ready if you wanna use only that with other classifiers.
— Reply to this email directly, view it on GitHub https://github.com/ijyliu/computer-vision-project/issues/70#issuecomment-2054207637, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQCGE4I37WH2BLA6R7VJULDY5MA2JAVCNFSM6AAAAABF5XEG66VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANJUGIYDONRTG4 . You are receiving this because you were assigned.Message ID: @.***>
See the graphs above
provides insights into what features are important (based off of autogluon, I think all are!)
we need to know what the best model is first!