Best accuracy parameters

[Nancy0577]

Hi,
I only can get around 30% of accuracy after 200 epochs. Could you please show me your parameters which can get over 50%?

Thanks

[DuaneNielsen]

How many runs have you tried?

From memory, convergence is highly dependent on the seed.

If it is not on track to converging early in training, just run it again.

[Nancy0577]

I had tried 10 runs. And what's your best accuracy that you have got before?

[DuaneNielsen]

From memory about 70% precision for MNIST. 1's and 7's were difficult to split right, 0's and 8's also...

…

On Tue, Dec 22, 2020 at 6:47 PM Nancy0577 ***@***.***> wrote: I had tried 10 runs. And what's your best accuracy that you have got before? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACUOYLHO6A7TV3NTY6SNTOLSWFK2PANCNFSM4VD7CPYQ> .

-- Duane Broadcom 480 760 1559

-- This electronic communication and the information and any files transmitted with it, or attached to it, are confidential and are intended solely for the use of the individual or entity to whom it is addressed and may contain information that is confidential, legally privileged, protected by privacy laws, or otherwise restricted from disclosure to anyone else. If you are not the intended recipient or the person responsible for delivering the e-mail to the intended recipient, you are hereby notified that any use, copying, distributing, dissemination, forwarding, printing, or copying of this e-mail is strictly prohibited. If you received this e-mail in error, please return the e-mail to the sender, delete it from your computer, and destroy any printed copy of it.

[Nancy0577]

How about your best accuracy on cifar10?

[DuaneNielsen]

Oh yeah, nowhere near that good on CIFAR. Probably about 30% or something like that. I think the background messes things up quite a bit, as the mutual information over the entire image gets dominated by the background. It works well on MNIST because the shape is the only source of information, and the shapes are quite distinct. Natural images would be quite a bit harder I think. Sorry :) heh.. I should have done a better writeup on this project.. but things move so fast, I'm often quick to rush to the next cool thing :)

…

On Tue, Dec 22, 2020 at 8:10 PM Nancy0577 ***@***.***> wrote: How about your best accuracy on cifar10? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACUOYLFGFM36PBFQU7AUCCDSWFUTPANCNFSM4VD7CPYQ> .

-- Duane Broadcom 480 760 1559

-- This electronic communication and the information and any files transmitted with it, or attached to it, are confidential and are intended solely for the use of the individual or entity to whom it is addressed and may contain information that is confidential, legally privileged, protected by privacy laws, or otherwise restricted from disclosure to anyone else. If you are not the intended recipient or the person responsible for delivering the e-mail to the intended recipient, you are hereby notified that any use, copying, distributing, dissemination, forwarding, printing, or copying of this e-mail is strictly prohibited. If you received this e-mail in error, please return the e-mail to the sender, delete it from your computer, and destroy any printed copy of it.

[Nancy0577]

This project is really good! And what's your clustering method? K-means? What is your accuracy? Purity, Precision or others?

[DuaneNielsen]

Here's the paper I was reproducing from. It's an information theory based algorithm called Invariant Information Clustering. https://arxiv.org/abs/1807.06653

…

On Wed, Dec 23, 2020 at 1:38 AM Nancy0577 ***@***.***> wrote: This project is really good! And what's your clustering method? K-means? What is your accuracy? Purity, Precision or others? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACUOYLEBYUEE47GW4OU77VLSWG3CVANCNFSM4VD7CPYQ> .

-- Duane Broadcom 480 760 1559

-- This electronic communication and the information and any files transmitted with it, or attached to it, are confidential and are intended solely for the use of the individual or entity to whom it is addressed and may contain information that is confidential, legally privileged, protected by privacy laws, or otherwise restricted from disclosure to anyone else. If you are not the intended recipient or the person responsible for delivering the e-mail to the intended recipient, you are hereby notified that any use, copying, distributing, dissemination, forwarding, printing, or copying of this e-mail is strictly prohibited. If you received this e-mail in error, please return the e-mail to the sender, delete it from your computer, and destroy any printed copy of it.