Thread: etiquette (sp?) of posting code

thats the problem, all i have is the bare minimum code for what i need to do. Basically, i got given 3 files, one .cpp file which contains main and handles the input from the user, one header file which contains the neural network class definition, and another .cpp file which contains the neuron class definition, and all functions associated with both the neuron class and the neural network class.

all i need to add is a Learn function to the neural network class, and thats all i have done (aswell as update the neuron and neural network class as needed). it *should* be working, as i get no compile errors and, unfortunately, it only breaks sometimes. this is why im so stumped!

thanks

alright, well you asked for it :P

Code:

void MLP::Learn(int numegs, double testset[], double lrate)
{
	int ct=0;
	int i=0;
	int egct=0;
	double sserr=0;
	double * outputs = new double [numegs];

	LinActNode ** Inputs = NULL;
	int NumInputs = 0;
	double * Weights = NULL;
	double * Invec = NULL;

	for (ct=0; ct<NumOutput; ct++)
	{ 
		i = NumInput+ct;	//use i rather than that expression, space saving
		//Nodes[i]->blah;

		Inputs = Nodes[i]->GetInputs();
		NumInputs = Nodes[i]->GetNumInputs();

		Weights = new double [NumInputs];

		Invec = new double [NumInputs];



		for (egct=0; egct<numegs; egct++)
		{
			cout << "Example " << egct << endl;

			//Load the input nodes with the example set egct
			Calculate (&testset[egct*3], outputs);

			//calculate the delta value of the output nod and store it
			Nodes[i]->SetDelta( testset[egct*3+2]  -  Nodes[i]->NodeOutput() );
			cout << "Delta set to: " << Nodes[i]->GetDelta() << endl;

			//get the current weights for the input to this node
			Weights = Nodes[i]->GetWeights();

			//Invec[] is the array of values getting passed into the node
			//invec 0 is the bias term
			Invec[0] = Weights[0];
			cout << "Invec[0] = " << Invec[0] << endl;
			for (int vecct=1; vecct<=NumInputs; vecct++)
			{
				//calculate invect as (output from all input nodes) * (weight associated)
				Invec[vecct] = Inputs[vecct-1] -> NodeOutput()  *  Weights[vecct];
				//cout << "Inputs[" << vecct-1 << "]->NodeOutput() = " <<  Inputs[vecct-1]->NodeOutput() << endl;
				cout << "Invec[" << vecct << "] = " << Invec[vecct] << endl;
			}

			for (int wct=0; wct<=NumInputs; wct++)
			{
				//update the weights for the node
				Weights[wct] = Weights[wct] + ( lrate * Invec[wct] * Nodes[i]->GetDelta() );
				cout << "Weights["<<wct<<"] = " << Weights[wct] << endl;
			}

			//set the weights for the node
			Nodes[i]->SetWeights(Weights);

			cout << endl;

		}


	}

}

i understand that it probably wont make much sense but any offers you could give as to why it spews out -1.#IND after i call that a few times would be appreciated (it accumulates to that value, numbers getting increasingly large or small).

thanks a lot.

Not seeing you delete weights and Invec, probably giving you a heap overflow after u call it a few times.

outputs need to be deleted as well. call delete [] outputs to delete an array.

cheers, will give it a go

back again;

Code:

			//set the weights for the node
			Nodes[i]->SetWeights(Weights);

			cout << endl;

		}

	}

	delete [] outputs;
	cout << "Deleted outputs[]" << endl;

}

thats the last few lines of the function. I tried deleting Weights, Invec, and Inputs, but they threw errors at me (runtime, rather than at compile etc).

Im fairly sure I dont need to delete Weights, as the function

Code:

Nodes[i]->GetWeights();

returns a pointer, so i have also removed the line which assigned memory to Weights (Weights = new double [NumInputs]. that is correct, yes? (same goes for Inputs).

now, i think Invec needs to be deleted as aside from outputs, it is the only other place where memory has been assigned. yet, when i try to call:

Code:

delete [] Invec;

at the end, i get the runtime error which spews some funny messages at me!

im going to attatch the files to this post so you can take a look if you like. i dont blame ya if you dont, but this is really helping me. i havent really got anyone to talk the problem through with at the moment (this doesnt have to be in for months, im getting a head start), and talking a problem out is awesome.

files are attatched, thanks again.

What compiler are you using? As you say you get "-1.#IND".
You should be able to find in the compiler documentation or on the internet all the possible cases when this output happens so you can possibly narrow it down even more.

c) posting the function and linking to a zip with the code

Personally, I won't download anything from a non-trusted source.

I see this in your code:

i = NumInput+ct;

but I don't see NumInput changing anywhere. You have another variable named NumInputs, are they separate variables or is that a mistake? Is Nodes[i] out of bounds?

enahs:

thanks for the advice, im using visual studio.net 2003. I have a feeling that i have a float which is getting ridiculously large or ridiculously small (ive seen it happening, so i know its this)... but crap knows how its happening. il keep a check on msdn for -1.#IND, though.


7stud:

dont worry about not downloading it, no worries =) the plain cpp, h, and txt files are above if you just dont want a zip, though.

NumInput and NumInputs are separate variables (not by my choice, but unfortunately i cant change the coursework skeleton ). Nodes[i] isnt going out of bounds - its not an instantaneous error and all the checks which i have placed indicate that codewise everything is perfectly in bounds and legal. im not so used to c++, though, so i could be well wrong =)

*sulk*

i think you have a fair point. I'll let you know how it goes.