Closed CIPop closed 6 years ago
With refinement. Looks like overfitting and potentially not picking up the right training vs validation samples. I'll check to see if I can upload the model to Github.
Hi Cris, did you figure out your overfitting suspicion? It also seems odd that your test accuracy would be higher than your train accuracy, no?
BTW I posted a connection string in the wiki you can use to upload models to blob storage.
Sent from Ninehttp://www.9folders.com/
From: Cristian Pop notifications@github.com Sent: Wednesday, July 25, 2018 4:31 PM To: Microsoft/arcticseals Cc: Peter Baer; Mention Subject: Re: [Microsoft/arcticseals] VGG16 + Adam LR=0.0001 (#17)
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FMicrosoft%2Farcticseals%2Fpull%2F17%23event-1753824647&data=02%7C01%7Cpbaer%40exchange.microsoft.com%7C235e3c6681d34bf1690208d5f286b8ca%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636681582801943625&sdata=89a4YJNgvhNdEnyLDPK99M01CKq1w%2FIIiZGLEnQGW2M%3D&reserved=0, or mute the threadhttps://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAHIdVL2K5vyrt1qHnDzQoI-HqR-aByJNks5uKP_FgaJpZM4Vgf5b&data=02%7C01%7Cpbaer%40exchange.microsoft.com%7C235e3c6681d34bf1690208d5f286b8ca%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636681582801953629&sdata=mWdN8eR2fMmrjN7ti3Jr7pRANnIeezqUvDpGai1L4uI%3D&reserved=0.
@pbaer sorry - lots of other work... getting through my email now. I didn't yet, no: I'd like to try out some other transfer models and even build one from scratch. It is indeed odd that is happening.
@pbaer Looks like there's something moving for the first time. Thanks to Eric for his idea, I'm seeing the net actually learning something: