I'm doing text classification using RNN. Text documents have varying length, therefore I would like to be able to specify sequence length when training the RNN. I noticed the sequence_length parameter in TensorFlowRNNClassifier() from skflow package, but do not know how to use it.
Suppose a batch of data consists of three documents of length 10, 20, and 30, what should the sequence_length parameter be? Just to be clear, I'm aware that we can set max_document_length to 30 and patch the first two documents with 0s, but I'd like to do dynamic sequences.
Thank you very much!
