CODEWITHSUNDEEP
tensorflowneural-networkrecurrent-neural-network
MatthewScarpino
MatthewScarpino

Reputation: 5936

TensorFlow's RNN units and cells

The constructor for many of the RNN classes (BasicRNNCell, LSTMCell, and so on) accepts an argument named num_units. This sets the number of units in the cell.

I thought this identified the number of elements the RNN should process in sequence. So if you want an RNN to process sequences of length N, you'd have N units per cell. Is this correct? What exactly is an RNN unit?

Upvotes: 0

Views: 1161

Answers (2)

Pegasus
Pegasus

Reputation: 21

here num_units refers to the number of units in LSTM(or rnn) cell.

num_units can be interpreted as the analogy of hidden layer from the feed forward neural network.The number of nodes in hidden layer of a feed forward neural network is equivalent to num_units number of LSTM units in a LSTM cell at every time step of the network.Following picture should clear any confusion- enter image description here

(cited from https://jasdeep06.github.io/posts/Understanding-LSTM-in-Tensorflow-MNIST/

Upvotes: 2

Lorenzo D'Arsiè
Lorenzo D'Arsiè

Reputation: 135

No, it's not correct.

num_units refers to the number of features your cells can represent. At each time step, you give an input of a certain size (that you are calling "the number of elements the RNN should process in sequence"). This is like the layer 0 of your neural network. This input is then processed into a hidden layer, with size num_units. This is also the size of the cell output.

What you call N, is set by the size of your inputs tensor. num_units is a hyperparameter of your model. The bigger it is, the more degrees of freedom your model has (more descriptive features).

Upvotes: 3

Related Questions