Ciro B Rosa
Ciro B Rosa
Hello, thanks for the response and follow up. Have managed to set my system up, thanks for that. Regards, Ciro. De: MartyG-RealSense ***@***.***> Enviada em: sexta-feira, 19 de agosto de...
> This is a great suggestion, but I'm unlikely to have time to implement it. Please consider a PR! It should be a straightforward change, I think Will upload a...
Have just uploaded new drawBoxes() function that accepts choice of bounding box color as kwarg, as discussed before. Some time ago we discussed about using a vector for choosing a...
Have made a quick-and-dirty fix, by creating a callback function, which can be executed after each epoch. Model is brought to CPU prior to saving its state. Have tested with...
> Indeed that error should be caught not thrown by the standard validation phase, but I can't see why the current code doesn't do that. Let me try and reproduce....
> Ah I see the issue is that early stopping throws a `CancelFittingException` but the epoch and step handlers only catch `CancelEpochException`/`CancelStepException`. See [here](https://github.com/FluxML/FluxTraining.jl/blob/8d24b5693b82b3c3e5a343c00662fc52a27f1952/src/training.jl#L91-L144). > > I guess the appropriate...
Cheers, Have implemented slightly different epoch training. Instead of using `fit!`, have used a pair of `epoch!` as follows: ``` epoch!(tr_learner, TrainingPhase(), trainset) epoch!(v_learner, ValidationPhase(), validset) ``` Learners for Training...
> I also saw that test and was confused about the behaviour. My problem with asking users to catch this is that CancelFittingException can be thrown for other, unrelated reasons....
Hello, same here.
Cheers, and sorry for long delay. To ease finding the root cause, have made my own dice_loss as follows: ``` function dice_loss(yhat::AbstractArray, y::AbstractArray) num = 2 * sum(yhat .* y)...