1.when i train RNN network in pytorch, the batch size is defined before the training.When i finish training,I want change the batch_size to predict the result of test set.How can i implement it?
2.when i train, i use GPU to train.when i predict the result, i wnat use the GPU, because my cpu’s memory is larger than the GPU’s.How can i implemet it?
I’m confused at them.Can you help me, please!
Thank you.
For 2, assuming you have a nn.Module somewhere that is your model and you’re trying to run on CPU for prediction, you can call model.cpu() and use that on the CPU.
From the docs, I don’t think there are restrictions on the batch_size for RNNs: http://pytorch.org/docs/master/nn.html?highlight=rnn#torch.nn.RNN .
I don’t know much about RNNs, but why can’t you just pass in an input with a different batch_size than the one you trained with?
If the batch_size is a problem, couldn’t you pad the rows of your input with zeros or random data?