Pytorch has been my main deep learning framework to work with. There is some part, however, that I felt could be improved. This has been answered by Pytorch Lightning [1].
William Falcon has laid out some of the core capabilities in Pytorch Lightning [2]. These features include structuring your codes to prepare the data, do training, validation, and testing, and logging with Tensorboard.
He has made an objective comparison between Pytorch Lightning, Pytorch Ignite, and fast.ai [4]. He highlighted that Ignite does not has a standard interface for every model, needs more line of code to train a model, is not directly integrated with Tensorboard, and does not has additional high-performance computing as Lightning does. While fast.ai has a higher learning curve than the other two and the use case may be different from Pytorch Lightning and Pytorch Ignite.

#python #deep-learning #data-science #artificial-intelligence #machine-learning