The Concatenate block merges 2 to 5 inputs into a single output that can be connected into other blocks.
This action is similar to creating feature sets, although you can use it anywhere inside your model and it is more adaptable to the data shape.
Number of inputs
You can concatenate between 2 and 5 inputs per Concatenate block. This number has to be selected when the block is created.
To change the number of inputs of an existing block, you will need to delete this block and create a new one.
Order of inputs
You can connect inputs to the concatenate block in any order you want. However, it is worth it to keep an ordering that is consistent:
With the meaning of your data, if there is any. For instance, when concatenating the left and right sides of an image, you should connect the left side to input 0 and the right side to input 1.
Between experiments, so that you can copy the weights that were trained in one experiment into another experiment.
The parameter Axis is the axis along which to concatenate.
Default is -1, which in
python means the last axis.
For 3D, if you want to merge the inputs vertically (1), horizontally (2) or depthwise (3).
Size of inputs
The size of all the inputs must be identical on each axis that is not the axis of concatenation. This is because you merge the inputs along the concatenation axis.
If you in 3D want to concatenate along the vertical axis, dimension 2, then all inputs must be identical along dimension 1 and 3.
Example use cases
Joining different features
Concatenating can be useful to merge features coming from different parts of the model.
See: Merge image features with tabular data as showcased in the tutorial Predict California house prices.
Joining image tiles
You can also use the Concatenate block to join together multiple images that are tiles of a bigger map.
Axis: Axis along which to concatenate, starting at 1 for the first axis.
Default: -1, meaning the last axis.