How could I get the "thought vector" using TensorLayer?

[munichong]

I am using the seq2seq model as an autoencoder. Given a test paragraph, I'd like to get the thought vector (using the terminology in the figure of README.md).