I created a deterministic convolutional neural network for classification, and then lifted it to a probabilistic network using
pyro.random_module(). I further tuned the learning rate as a hyper parameter during SVI optimization. While looping over SVI, I sampled the random network many times, e.g.,
sampled_models = [guide(None, None) for _ in range(num_model_samples)], to get many instances and evaluated the performance on a validation data set -- I wanted to keep the best probabilistic network. What can I do to save the best
In Pytorch, we can save the best network (call it myNet) by
What are the counterparts of
deepcopy, save, and load_state_dict in Pyro? Thanks.