-
Notifications
You must be signed in to change notification settings - Fork 146
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Use TrainingLoop in Autoencoder1D #677
Conversation
} | ||
var trainingLoop = TrainingLoop( | ||
training: dataset.training.map { $0.map { LabeledData(data: $0.data, label: $0.data) } }, | ||
validation: dataset.validation.map { LabeledData(data: $0.data, label: $0.data) }, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't think this is operating in the same way as the original model. For an autoencoder like this, we just want to train through the images once, then output a single input / output image pair at the end as the validation stage. We don't want to iterate through the validation set.
Also, in the original model the input / output images were captured before training, to show the random state at the very beginning. The test images themselves were different at each epoch, and they appear to be constant here.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for pointing out. Yes we'd better select different images every epoch to save for nicer manual validation.
Regarding iterating through the validation set, we'd better keep it as the standard way of computing stats?
e0a7dac
to
1202d54
Compare
1202d54
to
1910126
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The conversion makes sense, and is closer to the original implementation that what we have in the models right now, so this looks great to me.
Thanks for working this through.
No description provided.