NN-Tool

From Release 10 on, EBSILONProfessional allows an evaluation of neural networks created by means of the NN-Tool by Professor Bärmann (FH Gelsenkirchen).

Please note that EBSILON supports only the evaluation of networks. The creation of new networks and also the configuration and training of existing networks are not possible
within Ebsilon. That means that a network that was loaded one time will always yield the same output for the same input.

For loading as well as for the evaluation of a neural network an additional license is required.

The integration is effected via Component 93 (Kernel-Scripting).

Networks can either be imported into the model (and are thus part of the .ebs file) or they can be loaded on-the-fly.
The import is effected on the sheet "Basic Properties" of Component 93 in the Properties dialog.

The evaluation is carried out by means of the EbsScript unit "@NNTool". The core of the unit consists in the evaluation functions that expect as input an "nntoolreference" (or a path) and an array with input values in each case and write the results into the "output" array.

The "real" variants expect only double values, also for category values (ie 0 or 1 for each of the possible values, and thus correspondingly more input values), each of which has an "nntoolreference" (or path) and an array expect input values as input and write the result into the "output" array.

Name	Purpose	Arguments	Return value	Example
*Evaluation functions*
nnEvalReal	Evaluates a loaded neural network with input and output values of type REAL	Input: 1. CNNTOOLREFERENCE: Structure containing the ID reference code on the previously loaded network 2. const ARRAY OF REAL: Network input parameters Output: 3. ARRAY OF REAL: Network output parameters	NNTOOLRESULTENUM (details see Unit @NNTool)
nnEvalString	Evaluates a loaded neural network with input and output values of type STRING	Input: 1. const NNTOOLREFERENCE: Structure containing the ID reference code on the previously loaded network 2. const ARRAY OF STRING: Network input parameters Output: 3. ARRAY OF STRING: Network output parameters	NNTOOLRESULTENUM (details see Unit @NNTool)
nnEvalVariant	Evaluates a loaded neural network with input and output values of type VARIANT	Input: 1. const NNTOOLREFERENCE: Structure containing the ID reference code on the previously loaded network 2. const ARRAY OF VARIANT: Network input parameters Output: 3. ARRAY OF VARIANT: Network output parameters	NNTOOLRESULTENUM (details see Unit @NNTool)

Function for of single Evaluation

With each call, the net is loaded from the file (use nnLoad and nnEval [Real] with repeated evaluation for higher efficiency).

Name	Purpose	Arguments	Return value	Example
*Functions for a single evaluation*
nnLoadAndEvalReal	Loads and evaluates a neural network for the given path with input and output values of type REAL	Input: 1. STRING: Path to the neural network including filenames 2. Const ARRAY OF REAL: network input parameters Output: 3. ARRAY OF REAL: network output parameters	NNTOOLRESULTENUM (details see Unit @NNTool)
nnLoadAndEvalString	Loads and evaluates a neural network for the given path with input and output values of type STRING	Input: 1. STRING: Path to the neural network incl. file name 2. Const ARRAY OF STRING: input parameter to network Output: 3. ARRAY OF STRING: network output parameters	NNTOOLRESULTENUM (details see Unit @NNTool)
nnLoadAndEvalVariant	Loads and evaluates a neural network for the given path with input and output values of type Variant	Input: 1. STRING: Path to the neural network including filenames 2. Const ARRAY OF VARIANT: network input parameters Output: 3. ARRAY OF VARIANT: network output parameters	NNTOOLRESULTENUM (details see Unit @NNTool)

The nnLoad / nnUnload functions load or unload a network for the given path, whereas the
Function nnGetKsNNToolReference returns the references of an imported network.

Name	Purpose	Arguments	Return value	Example
nnLoad	Loads a neural network for the given path and returns the network reference (a variable of type NNTOOLRESULTENUM) to evaluate the network	Input: 1. STRING: Path to the neural network including filenames Output: 2. NNTOOLRESULTENUM: Result code for loading	NNTOOLREFERENCE - Structure containing the ID reference code on the loaded network
nnUnload	Discharges a previously charged neural network	Input NNTOOLREFERENCE: Structure containing the ID reference code on the loaded network	NNTOOLRESULTENUM (details see Unit @NNTool)
nnGetKsNNToolReference	Returns the reference of a network loaded in component 93	Input: 1. EBSCOMP93: Reference to Comp. 93 Output: 2. NNTOOLRESULTENUM: Result code of implementation	NNTOOLREFERENCE - Structure containing the ID reference code on the loaded network

Info-Functions:

Name	Purpose	Arguments	Return value	Example
nnGetPath	Returns the path of the loaded net as a string	Input: NNTOOLREFERENCE: Structure containing the ID reference code on the loaded network	STRING: the complete path incl. file name
nnGetInputCount	Returns the total number (real and string) of the input parameters of a network	Input: NNTOOLREFERENCE: Structure containing the ID reference code on the loaded network	INTEGER: Number of input parameters
nnGetInputRealCount	Returns the number of input parameters of a network of the real type	Input: NNTOOLREFERENCE: Structure containing the ID reference code on the loaded network	INTEGER:
nnGetOutputCount	Returns the total number (real and string) of the output parameters of a network	Input: NNTOOLREFERENCE: Structure containing the ID reference code on the loaded network	INTEGER: Number of output parameters
nnGetOutputRealCount	Returns the number of output parameters of a net of the type Real	Input: NNTOOLREFERENCE: Structure containing the ID reference code on the loaded network	INTEGER:
nnGetName	Returns the name of the parameter (input or output) for the specified index	Input: 1. NNTOOLREFERENCE: Structure containing the ID reference code on the loaded network 2. BOOLEAN: if true, then an input parameter, if false, then an output parameter 3. INTEGER: Index des Parameters	STRING: Name of the parameter
nnIsCategory	It returns true or false depending on whether the requested parameter is a discrete value or a value of type Real	Input: 1. NNTOOLREFERENCE: Structure containing the ID reference code on the loaded network 2. BOOLEAN: if true, then an input parameter, if false, then an output parameter 3. INTEGER: index of the parameter	BOOLEAN: true, if it is a discrete value (Category), otherwise false (Real)
nnGetCategoryNames	Returns the ARRAY of discrete parameter names	Input: 1. NNTOOLREFERENCE: Structure containing the ID reference code on the loaded network 2. BOOLEAN: if true, then an input parameter, if false, then an output parameter 3. INTEGER: index of the parameter	ARRAY OF STRING: Names of the discrete parameters

In the “examples” directory, there is an example model Optimierungsstudie_Shortcut_Modell_V10.ebs and the example network files Optimierung_100000_Datensaetze.wgt and Optimierung_100000_Datensaetze.dsc.

Impressum Disclaimer