Component: Neuron (Calculation)
| Author | Matrix TSL |
| Version | 1.3 |
| Category | Calculation |
Neuron component
Performs a single-neuron calculation of Z = W · X + B. Sums weight * input for each input, adds bias, then optionally applies an activation function (Linear, ReLU, Sigmoid, Tanh or Leaky ReLU). Use SetWeights / SetWeight and SetBias to configure, set inputs with SetInput or pass an array to Calculate / CalculateSingle. Use CalculateLoss or CalculateLossFromResult to measure error against a target value.
Detailed description
No detailed description exists yet for this component
Examples
No additional examples
Macro reference
Calculate
|
Calculate |
| Calculates Z = sum(Weight[i] * Input[i]) + Bias, then applies the selected activation. Returns the activated output and also stores it in Result. | |
 - FLOAT
|
InputArray |
| Array of input values (one per input) | |
| - FLOAT
|
Return |
CalculateLoss
|
|
CalculateLoss |
| Calculates loss between a predicted value and a target value using the Loss Type property. MSE = (Predicted - Target)^2. MAE = abs(Predicted - Target). Binary Cross Entropy = -(Target * log(Predicted) + (1 - Target) * log(1 - Predicted)). Stores the value in Loss and returns it. | |
| - FLOAT
|
Predicted |
| Predicted / output value | |
| - FLOAT
|
Target |
| Expected / target value | |
| - FLOAT
|
Return |
CalculateSingle
|
|
CalculateSingle |
| Convenience for a single-input neuron: Z = Weight[0] * X + Bias, then activation. Uses only input index 0. Ideal when Number of Inputs is 1. | |
| - FLOAT
|
X |
| Single input value | |
| - FLOAT
|
Return |
CalculateStored
|
|
CalculateStored |
| Same as Calculate, but uses the inputs previously stored with SetInput. Useful when filling inputs one at a time. | |
| - FLOAT
|
Return |
Initialise
|
|
Initialise |
| Clears all weights, inputs, bias, result and loss to 0.0. | |
 - VOID
|
Return |
SetBias
|
|
SetBias |
| Sets the bias offset constant B used in Z = W · X + B. | |
| - FLOAT
|
Value |
| Bias value | |
| - VOID
|
Return |
SetInput
|
|
SetInput |
| Stores a single input value for later use by CalculateStored. Index range: 0 to (Number of Inputs - 1). | |
 - BYTE
|
Index |
| Input index (0 based) | |
| - FLOAT
|
Value |
| Input value to store | |
| - VOID
|
Return |
SetWeight
|
|
SetWeight |
| Sets a single weight value by index. Index range: 0 to (Number of Inputs - 1). | |
| - BYTE
|
Index |
| Weight index (0 based) | |
| - FLOAT
|
Value |
| Weight value to store | |
| - VOID
|
Return |
SetWeights
|
|
SetWeights |
| Copies an array of weight values into the neuron. The array must contain at least Number of Inputs elements. | |
| - FLOAT
|
WeightArray |
| Array of weight values (one per input) | |
| - VOID
|
Return |
Property reference
|
Properties |
|
Number of Inputs |
| Number of inputs (and weights) used by the neuron. Range: 1 to 32. | |
|
Activation |
| Activation applied after Z = W · X + B. Linear returns Z unchanged. ReLU returns max(0, Z). Sigmoid returns 1 / (1 + exp(-Z)). Tanh returns (2 / (1 + exp(-2Z))) - 1. Leaky ReLU returns Z when Z > 0, otherwise Leak Slope * Z. | |
|
|
Loss Type |
| Error / loss function used by CalculateLoss and CalculateLossFromResult. MSE = (Predicted - Target)^2. MAE = abs(Predicted - Target). Binary Cross Entropy is intended for Sigmoid outputs in the range 0 to 1. | |
Component Source Code
Please click here to download the component source project: FC_Comp_Source_Neuron.fcfx
Please click here to view the component source code (Beta): FC_Comp_Source_Neuron.fcfx