BrunoBelucci

So, I have looked into it and found out that there were many types that needed to be correctly declared and there are also some incoherences in the declared types...

> np.float is indeed depreacated, but I'm not sure np.float32 is also deprecated. I am pretty sure float is np.float64. So you should keep np.float32 expressions since GPU usually works...

After some thought, I decided to change np.float to np.float32 following the assumption that we usually work with GPUs, so eps from np.float64 can be 0 if we use np.float32,...

I was facing the same issue when I wanted to use a custom scheduler and you have a solution that is pretty much what I have, the only thing that...

Honestly, I think that in this case, the cleanest solution is to add one more argument. We already have arguments for the trainer, optimizer, and scheduler. Introducing a different approach...

Hello, I just faced the same problem, started to use mlflow using a local mlrun directory because it was easier, accumulated ~ 10000 runs, realized that the UI (and searching...

Is there any plan or workaround to support int32?

Hello, Just stumbled into this while trying to figure out how I could change the optimizer / scheduler in the middle of the training. I have managed to do that...

Have you figured out how to use callbacks? My use case seems fairly simple, I just want to access the metrics and loss during training, but I still have not...

Hello, To the best of my knowledge, it is not currently possible to use ReduceLROnPlateau out-of-the-box with the existing implementation of neuralforecast. As mentioned in [this comment](https://github.com/Nixtla/neuralforecast/pull/998#issuecomment-2126773941), this would require...