How to code Neural Networks - Learning a Simple Function.

Опубликовано: 19 Сентябрь 2024
на канале: How to code Neural Networks

https://github.com/joaopauloschuler/n...

#NeuralAPI #GitHubRepository #caineuralapi

Learn the linear function 2*x + 3*y + 4
This example covers:
data preparation (the training data),
neural network architecture,
training and testing.

The Training Data
While training, we calculate inputs and expected outputs with:
```
vInputs[0] := (Random(10000) - 5000)/100; // Random number in the interval [-50,+50].
vInputs[1] := (Random(10000) - 5000)/100; // Random number in the interval [-50,+50].
vOutput[0] := 2*vInputs[0] - 3*vInputs[1] + 4; // 2x - 3y + 4
```
The actual data structures with input and output are defined with:
```
type
// Define the input and output types for training data
TBackFloatInput = array[0..1] of TNeuralFloat; // Input data for 2x - 3y + 4
TBackFloatOutput = array[0..0] of TNeuralFloat; // Expected output for 2x - 3y + 4
...
var
...
vInputs: TBackFloatInput;
vOutput: TBackFloatOutput;
```
Neural Network Architecture
The neural network consists of only 2 layers:
an input layer with two inputs (representing the two inputs of the OR operation)
and a single output neuron that provides the result. It uses a fully connected architecture without activation function.

The above is created with:
```
NN := TNNet.Create();
// Create the neural network layers
NN.AddLayer(TNNetInput.Create(2)); // Input layer with 2 neurons
NN.AddLayer(TNNetFullConnectLinear.Create(1)); // Single neuron layer connected to both inputs from the previous layer.
```

Training and Testing the Neural Network
The training and testing are done in a single code:
```
for EpochCnt := 1 to 100000 do
begin
vInputs[0] := (Random(10000) - 5000)/100; // Random number in the interval [-50,+50].
vInputs[1] := (Random(10000) - 5000)/100; // Random number in the interval [-50,+50].
vOutput[0] := 2*vInputs[0] - 3*vInputs[1] + 4; // 2x - 3y + 4
// Feed forward and backpropagation
NN.Compute(vInputs); // Perform feedforward computation
NN.GetOutput(pOutPut); // Get the output of the network
NN.Backpropagate(vOutput); // Perform backpropagation to adjust weights

if EpochCnt mod 5000 = 0 then
WriteLn(
EpochCnt:7, 'x',
' Output:', pOutPut.Raw[0]:5:2,' ',
' - Training/Desired Output:', vOutput[0]:5:2,' '
);
end;
```
In the above code, `vOutput` has the desired output while `pOutPut` has the output that was calculated by the neural network. In the output, you'll find:
```
90000x Output:39.17 - Training/Desired Output:39.17
95000x Output: 8.12 - Training/Desired Output: 8.12
100000x Output:110.15 - Training/Desired Output:110.15
```
As you can see above, the desired output is exactly the calculated output.

OUTLINE:
00:00 Introduction to Neural Networks
00:08 Untitled Chapter
00:14 Data Preparation for Neural Network
00:39 Neural Network Architecture
01:09 Training and Testing the Neural Network
01:29 Summary and Conclusion

#freepascal
#pascal
#delphi
#lazarus
#pascaldeeplearning
#pascalneuralnetworks
#programminglanguages
#deeplearning

Смотрите видео How to code Neural Networks - Learning a Simple Function. онлайн без регистрации, длительностью часов минут секунд в хорошем качестве. Это видео добавил пользователь How to code Neural Networks 19 Сентябрь 2024, не забудьте поделиться им ссылкой с друзьями и знакомыми, на нашем сайте его посмотрели 33 раз и оно понравилось 2 людям.