neuralnetwork how to choose the best model
I am using nftool to train my data (90 piece of data), and my training algorithm is "Bayesian Regularization" I am going to use different neurons to compare and find which one is the best model. Is there any appropriate step I need to follow? Do I need to retrain again and again until getting the good performance of model for each neurons?
See also questions close to this topic

MATLAB  Read Data from HEX using XCP and A2Lfile
I have the compiled version of an automotive ECU code in HEX format.
Using the Vehicle Network Toolbox, I want to read certain parameters from said HEX file by using the A2L file and XCP protocol.
I was able to import the data from the A2L file, using:
a2l_data = xcpA2L('file.a2l');
How can I use this to read data from the HEX file? Unfortunately the documentation on the Vehicle Network Toolbox is not as rich as I expected.

Elman neural network Learn multipleseries series
I start with neural networks, and I would like to know how I can do to have several input lists as an example of learning. In the documentation the example presented is the following :
[X,T] = simpleseries_dataset; net = elmannet(1:2,10); [Xs,Xi,Ai,Ts] = preparets(net,X,T); net = train(net,Xs,Ts,Xi,Ai); view(net) Y = net(Xs,Xi,Ai); perf = perform(net,Ts,Y)
where Xs and Ts are the input and the target, if i have serval series, should i repeat the learning for each series? thank you

Counting and taking time measurements Matlab
I have created a 2d vector which has a starting displacement of (0,5) it is attracted towards the center point (0,0) with its acceleration proportional to its distance from the center. The vectors displacement also has a random acceleration in any direction. The script currently plots the displacement of the vector on a 2d graph within a given time period, I now want to work out how to do the following:
Find out how many times the vectors distance from the center is equal to 1 (sqrt(x^2 + y^2)) within a given amount of time.
Measure the time at which the vectors displacement from the centre is equal to 1.
Any help or advice would be greatly appreciated!

using Artificial intelligence speech recognition naudio library
Has anyone ever managed to feed a neural network using the data extracted through the Naudio library?

Neuronal Network correct/training
Hello i need some help with my NN (i apologise for my meh english), i tested my code but i cant find the mistake.
Problem: My NN isnt learning much, i need like 500 Generations with 500 nn´s each to get it to emulate a XOR and it makes a mistakes every like 4th time i run it. I programmed it in this ugly style because i found out that with neuron classes and so on it was a bit slower, i may consider to change that back because i find it hard to change or read the code (dont know if it even works).
Goal: I wanted to make a simple Pong bot that takes pixels as input (thats why i skipped backprop.), like google deep mind ^^, but i cant even manage to get it done here. If you got some experience yourself or some advice i would appreciate it. And even with many research i couldnt find a compact article what NN i need to use to make it able to use a pixel as input (thought of white=1, black=1, red=0 but thats isnt the best way).
Info: Also what i noticed the strengh of the randomness has to be very high otherwise it will get stuck. Here is my ANN:
private float[][][] weights; //[Layer][Neuron][Weights to Neurons in previous Layer] public ANN(int input, int... neurons) { weights = new float[neurons.length][][]; for (int i = 0; i < neurons.length; i++) { weights[i] = new float[neurons[i]][input+1]; input = neurons[i]; } } public float[] calculate(float[] data) { //NN for (int i = 0; i < weights.length; i++) { //Layer data = layer(data, i); } return data; } private float[] layer(float[] data, int i) { float[] buffer = new float[weights[i].length]; for (int j = 0; j < weights[i].length; j++) { //Neuron buffer[j] = neuron(data, i, j); } return buffer; } private float neuron(float[] data, int i, int j) { float result = weights[i][j][0]; //Bias for (int k = 1; k < weights[i][j].length; k++) { //Weight result += data[k1] * weights[i][j][k]; } return sigmoid(result); } public static float sigmoid(float value) { //Is this Sigmoid even right!? return (float) ((2 / (1 + Math.exp(value)))1); } public void ajustWeights(int amount, float strengh, Random rdm) { for (int a = 0; a < amount; a++) { int i = rdm.nextInt(weights.length); int j = rdm.nextInt(weights[i].length); int k = rdm.nextInt(weights[i][j].length); weights[i][j][k] += (rdm.nextFloat()*21) * strengh; } }
Here the "Trainer":
private static final Random rdm = new Random(); private HashMap<ANN, Float> anns; public ANN train(ANN current, int gens, int amount) { anns = new HashMap<>(amount); for (int i = 0; i < gens; i++) { for (int j = 0; j < amount; j++) { ANN ann = new ANN(current); ann.ajustWeights(50, 10f, rdm); //Tests the ANN and gives it a score (ok its a negative error) float score = test(ann); anns.put(ann, score); } current = findBest(); float currentScore = anns.get(current); anns.clear(); anns.put(current, currentScore); } return current; } private ANN findBest() { float bestScore = Float.NEGATIVE_INFINITY; ANN bestANN = null; for (ANN key : anns.keySet()) { float value = anns.get(key); if (value < bestScore) continue; bestANN = key; bestScore = value; } return bestANN; }
Main code:
public static void main(String[] args) { ANN ann = new ANN(3, 8, 3); SingleTrainer trainer = new SingleTrainer(); ANN ret = trainer.train(ann, 5000, 5000); float[] data = new float[] {0,0,0}; for (int i = 0; i < 8; i++) { data = ret.calculate(data); //Maths.clamp(float[]) rounds all floats in the array to 1, 0 or 1 System.out.println(Arrays.toString(Maths.clamp(data))); } }
Thanks for ANY help.

Is it possible to implement neural network that outputs neural networks
So I started learning machine learning a little while ago, and as I see it, typical workflow for implementing machine learning algorithm is as follows:
 Reading in the data and cleaning it
 Exploring and understanding the input data
 Analyzing how best to present the data to the learning algorithm
 Choosing the right model and learning algorithm
 Measuring the performance correctly
And I was just wondering is it possible to somehow automate these mundane tasks, by implementing neural network that will learn to do it.