Background: I am trying to model multi-layered LSTM in tensorflow. I am using a common function to unroll the LSTM:
tf.nn.dynamic_rnn
Here I am using time_major = True, so my data has to be of format [max_time, batch_size, depth].
According to my understanding max_time is the time step of the series. My input is [224], I am passing into FC layer in the beginning to make it to size of labels
Question:
I am using datapipeline to get the labels in batch(32) of one_hot vector(length=70).
70 length vector corresponds to 1 time-step.
So how can I make the input to [32, 32 ,70]? Presently I have [32,70](batch_size,num_classes).
Please correct my understandings if wrong.
Can I just pass [1,32,70], so that tensorflow detects on its own, that each batch is of time step 1?
