ReplicateLayer

✖
`ReplicateLayer`

[Experimental]

✖

ReplicateLayer[n]

represents a net layer that takes an input of dimensions {d₁,d₂,…} and replicates it n times to produce an output of dimensions {n,d₁,d₂,…}.

✖

ReplicateLayer[{n₁,n₂,…,n_m}]

represents a net layer that takes an input of dimensions {d₁,d₂,…} and replicates it to produce an output of dimensions {n₁,n₂,…,n_m,d₁,d₂,…}.

✖

ReplicateLayer[dims,m]

replicates so that dims appears at position m in the list of output dimensions.

Details and Options

ReplicateLayer[n] is equivalent to ReplicateLayer[n,1].
For ReplicateLayer[n], the rank of the output array is one larger than the rank of the input array.
For ReplicateLayer[{n₁,…,n_m}], the rank of the output array is m larger than the rank of the input array.
Any of the n or n_i can have the value Automatic, leaving the exact output dimension to be inferred from its context in NetChain etc.
ReplicateLayer[…][input] explicitly computes the output from applying the layer to input.
ReplicateLayer[…][{input₁,input₂,…}] explicitly computes outputs for each of the input_i.
When given a NumericArray as input, the output will be a NumericArray.
ReplicateLayer exposes the following ports for use in NetGraph etc.:
"Input" an array of arbitrary rank

"Output" an array of greater rank than the input
ReplicateLayer is typically used inside NetChain, NetGraph, etc.
ReplicateLayer normally infers the dimensions of its input from its context in NetChain etc. To specify the dimensions explicitly as {n₁,n₂,…}, use ReplicateLayer["Input"->{n₁,n₂,…}].
Options[ReplicateLayer] gives the list of default options to construct the layer. Options[ReplicateLayer[…]] gives the list of default options to evaluate the layer on some data.
Information[ReplicateLayer[…]] gives a report about the layer.
Information[ReplicateLayer[…],prop] gives the value of the property prop of ReplicateLayer[…]. Possible properties are the same as for NetGraph.

Examples

open allclose all

Basic Examples (2)Summary of the most common use cases

Create a ReplicateLayer that replicates its input 4 times:

In[1]:=1

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-kq71xu

Out[1]=1

Create a ReplicateLayer that replicates its input 3 times:

In[1]:=1

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-zi5xha

Out[1]=1

Apply the layer to data:

In[2]:=2

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-7u8hba

Out[2]=2

Scope (6)Survey of the scope of standard use cases

Arguments (2)

Create a ReplicateLayer that replicates its input 3×4 times:

In[176]:=176

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-yb04cp

Out[176]=176

Apply the layer to a vector:

In[2]:=2

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-ix0v19

Create a ReplicateLayer that replicates the elements of a vector into rows of length 3:

In[1]:=1

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-lwd51p

Out[1]=1

Apply the layer to an input:

In[2]:=2

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-6afzpp

The elements can be replicated into columns instead:

In[3]:=3

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-sgr2fa

Out[3]=3

In[4]:=4

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-vqfl0w

Ports (4)

Create a ReplicateLayer that takes an input of specific size:

In[179]:=179

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-06s1bp

Out[179]=179

Apply the layer to an input:

In[2]:=2

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-eqiagw

Create a NetGraph containing a ReplicateLayer that automatically broadcasts a vector to a matrix of given size:

In[1]:=1

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-4meoer

Out[1]=1

This NetGraph is now valid for multiple input dimensions:

In[2]:=2

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-pwdh21

Out[2]=2

In[3]:=3

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-pjqe8s

Out[3]=3

Create a ReplicateLayer that replicates a single number to form a vector:

In[1]:=1

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-028rfm

Out[1]=1

Apply the layer to a number:

In[2]:=2

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-1d476e

Out[2]=2

The layer threads across a batch of inputs:

In[3]:=3

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-kb56t7

Out[3]=3

Create a NetChain that makes four copies of an image by replicating and then flattening both spatial dimensions in turn:

In[1]:=1

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-m2efma

Out[1]=1

In[2]:=2

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-j58jfb

Out[2]=2

Possible Issues (2)Common pitfalls and unexpected behavior

ReplicateLayer cannot accept non-numeric inputs:

In[188]:=188

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-ggaxyd

Out[189]=189

In[190]:=190

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-56whoo

Out[190]=190

ReplicateLayer[dims,pos] will fail when pos is more than one greater than the rank of the input:

In[1]:=1

✖

https://wolfram.com/xid/0bh1n2o11wfwrm-s85t60

Out[1]=1

Top