PyTorch model training parameters

Is there a method available to find the best training parameters such as Epoch, learning rate, and batch size for a deep learning model? How do people go about find those values, please?

I think usually these hyperparameters are determined heuristically, i.e. by testing a lot of values or finetuning “good known” values. There are some approaches to automate this, such as FasiAI’s learning rate finder. Also, the general approach would be “learning to learn” or meta learning and is a current area of research.

1 Like

Thanks very much, That really useful.