WOLFRAM

Wolfram Language & System Documentation Center

NetPrepend[net,layer]

prepends a layer or a net onto a NetChain, a layer or a NetGraph with one input port.

NetPrepend[net,"name"layer]

appends a layer with a given name.

NetPrepend[net,{layer1,layer2,…}]

prepends several layers or nets.

Details and Options
Details and Options Details and Options
Examples  
Basic Examples  
Scope  
Options  
"Input"  
"Output"  
Possible Issues  
See Also
Tech Notes
Related Guides
History
Cite this Page

NetPrepend

NetPrepend[net,layer]

prepends a layer or a net onto a NetChain, a layer or a NetGraph with one input port.

NetPrepend[net,"name"layer]

appends a layer with a given name.

NetPrepend[net,{layer1,layer2,…}]

prepends several layers or nets.

Details and Options

  • The output of the prepended layer must be compatible with the input of the first layer in the chain.
  • Any NetEncoder attached to the end of the chain will be removed.
  • NetPrepend takes the following options:
  • "Input"Automatictype for the input of the resulting chain
    "Output"Automatictype for the output of the resulting chain
  • Options "Input" and "Output" support the normal type specifications allowed by NetChain, including NetEncoder and NetDecoder.

Examples

open all close all

Basic Examples  (3)

Prepend a layer to the start of a chain:

Wolfram Language code: NetPrepend[NetChain[{4, SoftmaxLayer[]}], ElementwiseLayer[Tanh]]

Prepend a named layer to the start of a chain:

Wolfram Language code: NetPrepend[NetChain[<|"Linear" -> 4, "Softmax" -> SoftmaxLayer[]|>], "Tanh" -> ElementwiseLayer[Tanh]]

Prepend several layers to the start of a chain:

Wolfram Language code: NetPrepend[NetChain[{Tanh, 4, SoftmaxLayer[]}], {UnitVectorLayer[], LinearLayer[4]}]

Scope  (3)

Prepend a layer and attach a NetEncoder:

Wolfram Language code: NetPrepend[NetChain[{4, SoftmaxLayer[]}], UnitVectorLayer[], "Input" -> NetEncoder[{"Class", {"a", "b"}}]]

Create a NetChain with NetPrepend:

Wolfram Language code: NetPrepend[LinearLayer[], Ramp]

Prepend a layer to a NetGraph with one input:

Wolfram Language code: ng = NetGraph[{1, 2, 3}, {1 -> {2, 3}}]
Wolfram Language code: NetPrepend[ng, LinearLayer[]]

Prepend a layer with a given name:

Wolfram Language code: NetPrepend[ng, "Foo" -> LinearLayer[]]

Options  (6)

"Input"  (4)

Set the input shape of the net:

Wolfram Language code: NetPrepend[NetChain[{ConvolutionLayer[3, 3], Tanh}], ElementwiseLayer[Tanh], "Input" -> {3, 256, 256}]

Set a NetEncoder for the net:

Wolfram Language code: NetPrepend[NetChain[{ConvolutionLayer[3, 3], Tanh}], ElementwiseLayer[Tanh], "Input" -> NetEncoder[{"Image", {256, 256}}]]

Using this option is equivalent to using NetReplacePart on the returned net:

Wolfram Language code: net = NetPrepend[NetChain[{ConvolutionLayer[3, 3], Tanh}], ElementwiseLayer[Tanh]]
Wolfram Language code: NetReplacePart[net, "Input" -> NetEncoder[{"Image", {256, 256}}]]

An existing NetEncoder can be added back to the returned net using this option, provided it is compatible with the new initial layer:

Wolfram Language code: enc = NetEncoder["Image"]
Wolfram Language code: net = NetChain[{ConvolutionLayer[3, 3], Tanh}, "Input" -> enc]
Wolfram Language code: NetPrepend[net, ElementwiseLayer[Tanh]]
Wolfram Language code: NetPrepend[net, ElementwiseLayer[Tanh], "Input" -> enc]

"Output"  (2)

Set a NetDecoder for the net:

Wolfram Language code: NetPrepend[NetChain[{ConvolutionLayer[3, 3], Tanh}], ElementwiseLayer[Tanh], "Output" -> NetDecoder["Image"]]

Using this option is equivalent to using NetReplacePart on the returned net:

Wolfram Language code: net = NetPrepend[NetChain[{ConvolutionLayer[3, 3], Tanh}], ElementwiseLayer[Tanh]]
Wolfram Language code: NetReplacePart[net, "Output" -> NetDecoder["Image"]]

Possible Issues  (3)

The output of the prepended layer must be compatible with the input of the first layer in the chain:

Wolfram Language code: NetPrepend[NetChain[{PoolingLayer[{3}], Ramp}], ConvolutionLayer[3, {2, 2}]]

A network with several outputs cannot be prepended:

Wolfram Language code: NetPrepend[NetChain[{1, 2, 3}], NetGraph[{1, 2}, {}]]

A layer cannot be prepended to a network with more than one input:

Wolfram Language code: ng = NetGraph[{1, 2, Plus}, {{NetPort["Input1"] -> 1, NetPort["Input2"] -> 2} -> 3}]
Wolfram Language code: NetPrepend[ng, LinearLayer[]]
Wolfram Research (2018), NetPrepend, Wolfram Language function, https://reference.wolfram.com/language/ref/NetPrepend.html (updated 2021).

Text

Wolfram Research (2018), NetPrepend, Wolfram Language function, https://reference.wolfram.com/language/ref/NetPrepend.html (updated 2021).

CMS

Wolfram Language. 2018. "NetPrepend." Wolfram Language & System Documentation Center. Wolfram Research. Last Modified 2021. https://reference.wolfram.com/language/ref/NetPrepend.html.

APA

Wolfram Language. (2018). NetPrepend. Wolfram Language & System Documentation Center. Retrieved from https://reference.wolfram.com/language/ref/NetPrepend.html

BibTeX

@misc{reference.wolfram_2026_netprepend, author="Wolfram Research", title="{NetPrepend}", year="2021", howpublished="\url{https://reference.wolfram.com/language/ref/NetPrepend.html}", note=[Accessed: 12-June-2026]}

BibLaTeX

@online{reference.wolfram_2026_netprepend, organization={Wolfram Research}, title={NetPrepend}, year={2021}, url={https://reference.wolfram.com/language/ref/NetPrepend.html}, note=[Accessed: 12-June-2026]}