The Operations components provide common math operations that can be applied to layers in a neural network, and more generally, to tensors and matrices.


Performs an Argmax function that computes the maximum values for a set of inputs that would result in a maximum output.

This is typically the last layer of the model in cases where you only want a single answer rather than a distribution (e.g., for classification).


  • Dimension: specifies which axis contains the inputs on which to compute Argmax.


Combines two or more layers within a neural network such as when building skip connections for ResNets.


  • Number of inputs: specifies the number of input components to merge. Adjusting this sets the number of input sockets available in the component to which other components can be dragged and connected to.

  • Operation: specifies the type of merge to perform. Can be set to:

    • Concatenate: concatenates the values from each input. You can also set which dimension you want to concatenate on, where -1 defaults to the last dimension.

    • Subtraction: subtracts the values from each input.

    • Addition: adds the values from each input.

    • Multiplication: multiples the values from each input.

    • Division: divides the values from each input.


Switches between two different inputs and sends only the output from one of them at a time. This component "swaps" every other time (i.e., first component, second component, first component etc.).

This can be useful for models involving multiple inputs such as GANs where image data from two sources are connected to the component's input sockets, and the output is connected to a Dense layer.


Performs a Softmax function to normalize the sum of the inputs.

This is typically the last layer of the model in cases where you want all your outputs to sum up to 1 (e.g., for classification).

Last updated