Confused about how to setup a NN

2

to answer your three questions:

  1. We chose to make the examples to be best practices. We dont suggest users to use sequential except for basic convenience. Sequential becomes inflexible very quickly.

  2. You can use this recently added function http://pytorch.org/docs/nn.html#torch.nn.Module.named_parameters to filter out just the ELU parameters and not send them to the optimizer.

  3. you can use the environment variable CUDA_VISIBLE_DEVICES=“device_id” to control which GPU to use. For example CUDA_VISIBLE_DEVICES=2 python main.py # uses GPU-3

1 Like