SNN

Complete Two Layer SNN. More...

Data Fields
I32	IL
	set by snn_init().
I32	H1
	set by snn_init().
I32	H2
	set by snn_init().
I32	OL
	set by snn_init().
F32 *	iLTmp
	can also keep bias neuron, allocated by snn_init(), freed by snn_deinit().
F32 *	oLTmp
	can also keep bias neuron, allocated by snn_init(), freed by snn_deinit().
F32 *	h1
	allocated by snn_init(), freed by snn_deinit().
F32 *	h2
	allocated by snn_init(), freed by snn_deinit().
F32 *	w1
	allocated by snn_init(), freed by snn_deinit().
F32 *	w2
	allocated by snn_init(), freed by snn_deinit().
F32 *	w3
	allocated by snn_init(), freed by snn_deinit().
F32 *	delta1
	allocated by snn_init(), freed by snn_deinit().
F32 *	delta2
	allocated by snn_init(), freed by snn_deinit().
F32 *	delta3
	allocated by snn_init(), freed by snn_deinit().
F32 *	e3
	allocated by snn_init(), freed by snn_deinit().
F32 *	e2
	allocated by snn_init(), freed by snn_deinit().
F32 *	e1
	allocated by snn_init(), freed by snn_deinit().
I32	random_seed
	set by snn_train_init().
F32	fmax_weight
	set by snn_train_init().
F32	learning_rate
	set by snn_train_init().
F32	momentum
	set by snn_train_init().
F32(*	sigmoid_func )(F32)
	set by snn_init().
F32(*	differentiation_func )(F32)
	set by snn_init().
F32(*	confidence_func )(I32, F32 *)
	first argument is the count of Output Neurons, float argument is output from SNN, returns confidence. Set by snn_init().
F32(*	deviation_func )(I32, F32 *, F32 *)
	first argument is the count of Output Neurons, first float argument is output from SNN, second float argument is should-be from dataset, returns deviation. Set by snn_train_init().
I32(*	decision_func )(I32, F32 *)
	decider function returns index based on the given output Layer values. Set by snn_init().

Detailed Description

There is the last Neuron special in every Layer:

• Input Layer: Last Neuron (Index: IL) contains BIAS.
• Hidden Layers: Last Neuron (Index: H1 or H2) always equals 1 at start.

  See also

  [http://cs.uni-muenster.de/Professoren/Lippe/lehre/skripte/wwwnnscript/backprop.html#bias]

  Parameters

  IL	Count of Inputs.
  H1	Count of Neurons in the Hidden Layer 1.
  H2	Count of Neurons in the Hidden Layer 2.
  OL	Count of Outputs.
  w1	Weights between the Inputs and the Neurons in the Hidden Layer 1: weight[From Input i to Layer 1 Neuron j] = *(w1 + j * (IL+1) + i).
  w2	Weights between the Neurons in the Hidden Layer 1 to Layer 2: weight[From Layer 1 Neuron i to Layer 2 Neuron j] = *(w2 + j * (IL+1) + i).
  w3	Weights between the Neurons in the Hidden Layer 2 to Output: weight[From Layer 2 Neuron i to Output j ] = *(w3 + j * (IL+1) + i).
  h1	Buffers the h1 neuron's Output Value at the Processing Step between calculating from Input to Layer 1 and calculating from Layer 1 to Layer 2: [neuron k] = 1/(1+exp(sum_{i} w1_{k,i} * inputval_i)).
  h2	Buffers the h2 neuron's Output Value at the Processing Step between calculating from Layer 1 to Layer 2 and calculating from Layer 2 to Output: [neuron k] = 1/(1+exp(sum_{i} w2_{k,i} * h1_i)).
  e3	At learning: The deviation of the correct output signal from the value of the Neuron at the Hidden Layer 2: difference[correct Output, calculated Output] * non-standard weighting with correctO * (1 - correctO)
  e2	At learning: The deviation of the backpropagated signal from the forward propagated signal.