So i understand that RNN has hidden units within each RNN cell, as shown in the visualization of the second answer in the link:
What is num_units in tensorflow BasicLSTMCell?
However, I've been wondering, does increasing the number of hidden units within each cell improve the accuracy of the model?
I've been wondering, does increasing the number of hidden units within each cell improve the accuracy of the model?
In general, it would be true to say it increases the capacity of the model. That is, the capacity of the model to learn more complex correlations given enough data and training time.
Whether or not this leads to an increase in accuracy is part of the art! If you have too complex a model for your dataset you could end up overfitted, short of sufficient training data, or waiting for hell to freeze over before enough training computations can be completed. It's a balance and too subtle a question for a more specific answer!
