CatenateLayer

✖
`CatenateLayer`

[Experimental]

✖

CatenateLayer[]

represents a net layer that takes a list of input arrays and catenates them.

✖

CatenateLayer[n]

represents a net layer that takes a list of input arrays and catenates them at level n.

Details and Options

CatenateLayer[] is equivalent to CatenateLayer[1].
CatenateLayer is typically used inside NetGraph.
CatenateLayer[…][{array₁,array₂,…}] explicitly computes the output given a list of array_i.
For a list of input arrays {array₁,array₂,…}, the array_i must be of compatible dimensions. If necessary, the array_i are replicated as appropriate in order to make them all the same rank.
If given, the level n is relative to the first dimension of the array_i with smallest rank.
Option InputPorts can be used to specify the number, names or shapes of input ports. If InputPorts is not specified, the number of ports and their shapes are inferred from NetChain or NetGraph connectivity.
Besides input ports, CatenateLayer exposes the following ports for use in NetGraph etc.:
"Output" an array
Within a NetGraph, a CatenateLayer can be connected using a single edge of the form {src₁,src₂,…}catlayer, where catlayer is the name or index of the CatenateLayer, or as multiple separate edges given in the corresponding order, as src₁catlayer,src₂catlayer,…,src_ncatlayer.
When given a NumericArray as input, the output will be a NumericArray.
Options[CatenateLayer] gives the list of default options to construct the layer. Options[CatenateLayer[…]] gives the list of default options to evaluate the layer on some data.
Information[CatenateLayer[…]] gives a report about the layer.
Information[CatenateLayer[…],prop] gives the value of the property prop of CatenateLayer[…]. Possible properties are the same as for NetGraph.

Examples

open allclose all

Basic Examples (3)Summary of the most common use cases

Create a CatenateLayer:

In[1]:=1

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-lqs1z4

Out[1]=1

Create a CatenateLayer that takes two vectors as input and returns a vector:

In[1]:=1

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-m3tu8d

Out[1]=1

Apply this layer to a list of vectors:

In[2]:=2

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-i84nc7

Out[2]=2

Construct a NetGraph with a CatenateLayer, specifying input sizes:

In[1]:=1

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-jwzxw3

Out[1]=1

Apply the net to input data:

In[2]:=2

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-f929t7

Out[2]=2

Scope (4)Survey of the scope of standard use cases

Create a CatenateLayer that catenates the rows of two matrices:

In[1]:=1

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-bg20s9

Out[1]=1

Apply the layer to two matrices:

In[2]:=2

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-dky1z7

Create a CatenateLayer that catenates a matrix and a vector by repeating the vector:

In[1]:=1

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-svrk6j

Out[1]=1

Apply the layer to an input matrix and vector:

In[2]:=2

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-4iuohx

Out[2]=2

Construct a NetGraph with a CatenateLayer that takes two matrices representing sequences of vectors as input and catenates the vectors:

In[1]:=1

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-ce8zcd

Out[1]=1

Apply the net to two sequences to create a single sequence of vectors:

In[2]:=2

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-829k7y

Construct a NetGraph with a CatenateLayer that takes two images and catenates them horizontally:

In[1]:=1

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-l1m2a1

Out[1]=1

Apply the net to two images to create a single image:

In[2]:=2

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-1zlsfo

Out[2]=2

Properties & Relations (1)Properties of the function, and connections to other functions

When the inputs are all the same rank, CatenateLayer[1] computes:

In[1]:=1

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-tqmrp1

Evaluate on some data:

In[2]:=2

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-cfaack

Out[2]=2

Possible Issues (4)Common pitfalls and unexpected behavior

When catenating arrays of equal rank, the off-level dimensions must be equal:

In[1]:=1

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-nix1fw

Out[1]=1

In[2]:=2

✖

https://wolfram.com/xid/0b7esxf2bjd4uysi-g9mir7

Out[2]=2

When catenating arrays of different rank, the dimensions of lower-rank arrays should be suffixes of the dimensions of higher-rank arrays, except for the dimension in each array that corresponds to the level:

In[1]:=1