Create a Layer from Wolfram Language Code

Write a function to compute the distance between two vectors:

In[1]:=

f=Function[Module[{squarediff,mean},squarediff=(#a-#b)^2;mean=Sqrt@Mean[squarediff]]];

Apply the function to a set of vectors:

In[2]:=

f["a"{1.,2.},"b"{2.,4.}]

Out[2]=

1.58114

Use

FunctionLayer

to convert the function to a custom neural network layer:

In[3]:=

net=FunctionLayer[f]

Out[3]=

FunctionLayer



a :	array
b :	array
Output :	array



Check that the layer gives the same values as the function:

In[4]:=

net["a"{1.,2.},"b"{2.,4.}]

Out[4]=

1.58114

Next use

FunctionLayer

to create the

NetGraph

corresponding to different Wolfram Language symbols. For example, represent commonly used distances as network layers:

In[5]:=

AssociationMap[NetGraph@*FunctionLayer@*Apply,{EuclideanDistance,ManhattanDistance,ChessboardDistance,HammingDistance,BrayCurtisDistance}]

Out[5]=

EuclideanDistanceNetGraph

Number of inputs:	2
Output port:	real

,ManhattanDistanceNetGraph

Number of inputs:	2
Output port:	real

,ChessboardDistanceNetGraph

Number of inputs:	2
Output port:	real

,HammingDistanceNetGraph

Number of inputs:	2
Output port:	non-negative integer

,BrayCurtisDistanceNetGraph

Number of inputs:	2
Output port:	real



Create a normalization layer:

In[6]:=

normalizationLayer=FunctionLayer[Normalize]

Out[6]=

FunctionLayer

Input :	array
Output :	array



Check the corresponding

NetGraph

:

In[7]:=

normalizationLayer//NetGraph

Out[7]=

NetGraph

InputPort
Input :	array
OutputPort
Output :	array



Wolfram Language Example Repository

See Also

Publisher Information

Create a Layer from Wolfram Language Code

See Also

Related Symbols

Publisher Information