Variable-length LSTM in a mini-batch

[qqaatw]

Environment:

1. Python 3.6.3
2. Mxnet-cu92
3. Cuda 9.2 with cuDNN

I'm now proceeding a text processing networks, and have variable length inputs (time-steps) in a batch, I have already padded zeros if the sequence length is not enough to the max sequence length.

Is there a way to mask padded zeros and avoid it being updated?

p.s
I saw this PR before, but I'm not sure that padded zeros would be updated or not.

[mxnet-label-bot]

Hey, this is the MXNet Label Bot.
Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it.
Here are my recommended labels: Feature

[frankfliu]

@mxnet-label-bot add [question, operator]