Do you have an example of an LSTM from streaming data with more than one feature of numerical data. Also the sequences all have different lengths, is there also an example using
tf.contrib.training.bucket_by_sequence_length
Using that function allows one to stratify minibatches into groups (“buckets”) and if you use dynamic_pad=True, the minibatches will group similarly sized sequences for efficient training via dynamic_rnn. However, there aren’t many examples on how to use this function and the only examples regarding LSTM RNN is related to text sequences and it’s univariate. There’s no numerical streaming data with more than one feature examples?