What's the different between mx.model.load_checkpoint and mx.module.Module.load?


#1

Hi,

Seems that there are two ways to load model from previous checkpoints for prediction or continue to train,

mxnet.model.load_checkpoint(checkpoint, epoch);

or

mxnet.module.Module.load(checkpoint, epoch, …);

May I ask what’s the difference? I want to train the model for several iteration and save the model by checkpoints for each of them. And then later load model from checkpoint, make prediction and evaluate for each iteration. Which is correct way please?


#2

Module.load internally invokes load_checkpoint.
Module.load returns a fully initialized Module object using which you can either continue training or run predictions.