Values meaning :

 

value

In/Out

level(s)

action

ipNInput

IO

3/0

Model input number.

ipNOutput

O

0

Model output number

ipNOrder

O

2

Model order

ipCanLearn

O

1

Training capacity (0: no; 1: yes)

ipDimension

O

0

Parameter vector size

ipNbLayer

O

0

Model layer number

ipClass

O

3

Model pilot class index(note 1)

ipNHidden

O

0

Hidden nodes number

ipBootStrap

IO

3

Bootstrap number

ipStatus

O

0

Model status (0 ou error code)

ipNNeuron

O

0

Model node number

ipNModelInput

O

0

Model input number before eventual size reduction

ipNSynapse

O

0

Model parameter number.

ipNData

O

0

Training data number de

ipTrainCost

IO

2/0

Cost function index (note 2)

ipEndTrain

O

0

Training stop reason (note 3)

ipDataLoaded

O

0

Training data are loaded (0: no; 1: yes)

ipIsLin

O

0

Parameters linear model (0: no; 1: yes)

ipCommentCount

O

0

Model stored comments number

ipModelCount

O

3

Model count in multi models

ipNFuncAct

O

0

Available activation function number.

ipNFuncCost

O

0

Available cost function number.

ipNFuncProx

O

3

Available proximity function number (Kohonen models)(note 4)

ipNFuncDecay

O

3

Available decreasing function number (Kohonen models)(note 5)

ipValidHandle

O

0

Valid handle (0: no; 1: yes)

ipBin

IO

2/2

Binary storing forcing.

ipNTrainModel

O

2

Stored models stored number for multi training.

ipBootStrapType

IO

2/2

Bootstrap type (note 6)

ipTrainingAlgorithm

I

0/0

Training algorithm (note 7)

ipSelectedOutput

IO

0/0

Selected output (default 0)

ipProximity

IO

0/0

Proximity function index (Kohonen models)

ipDecay

IO

0/0

Decay function index (Kohonen models)

ipProximityDecay

IO

0/0

Distance decay function index (Kohonen models)

ipReverseMode

IO

0/0

Reverse mode (note 8).

ipBestIndex

O

0

Best training index, following the criterium

ipCriterium

IO

0/0

Model selection criterium. (note 9)

ipMultiTrainCount

O

2

Train count for multi train procedure

ipNTest

IO

0/0

Test set size

ipTestGroup

IO

0/0

Grouped test data

ipSaveTemp

IO

1/1

Do save on disk the temporary results

ipLoadTrainedModel

O

1

Load the trained model of index.

       

 

Notes :

1. Model pilot class index :

               1        Usage

               2        Static training

               3        Dynamic training

               4        Reverse mode

               11        Linear symetric matrix

               12        Linear square matrix

               13        Linear matrix

               14        Linear vector

 

2. Available cost functions index :

               0        Null cost

               1        Error square

               2        Delta values square

               3        Lessed delta values square

               4        Weighted squared error

               5        Badly classified

               6        bascule

               7        Relative error square

               8        Gaussian error square

               9        Crossed entropy

               10        Exponential square error

 

3. Training stop reason :

               0        Current training

               1        Reached objective

               2        Epoch number reached

               3        Computation accuracy

               4        Strong parameter

               5        Computation error

               6        User ask

               7        Best validation

 

4. Available proximity function index :

               0        1/(1 + Distance)

               1        1/sqrt(1 + Distance)

               2        Exp(-Distance)

 

5. Available decreasing function index :

               0        (Ratio < 1)

               1        Exp(-Ratio/2)

 

6. Bootstrap type :

               0:        nothing (default)

               1:        Standard

               2:        Residual.

 

7: Training algorithm :

               0:        Gradient; (default)

               1:        BFGS;

               2:        Levenberg-Marquardt.

 

8. Reverse modes :

               0:        icNone                nothing

               1:        icMinimisation        output minimization

               2:        icMaximisation        output maximization

               3:        icReverse        target-output delta minimization

               4:        icReverseSqrMin target-output delta minimization, while minimizing the squared inputs.

               5:        icReverseSqrDeltaMin        target-output delta minimization, while minimizing the inputs modifications.

 

9. Model selection criteria

 

Value

Name

Meaning

0

Tcr_Cost

Training cost

1

Tcr_Test

Test cost

2

Tcr_Press

PRESS

3

Tcr_Mu

Homogeneity

4

Tcr_StdDevBiasLess

Biasless standard deviation

5

Tcr_R2

R2

6

Tcr_R2Adjust

Adjusted R2

7

Tcr_Correl

Correlation

8

Tcr_DeltaRank

Jacobien rank defect

9

Tcr_Determinant

Determinant

10

Tcr_EigenMin

Higher Dispersion Matrix Eigen value

11

Tcr_EigenMax

Lower Dispersion Matrix Eigen value

12

Tcr_Conditionning

Ratio of the two upper

13

Tcr_Trace

Dispersion Matrix Trace

14

Tcr_CorrelMax

Maximum of parameters absolute correlation

15

Tcr_CorrelMean

Mean of parameters absolute correlation

 

 

 


NETRAL Neuro Developer Kit version 7.0