NARMA prediction task parameter exploration

Initial network setup

Rendered by QuickLaTeX.com

Spectral radius

argmin NRMSE=0.95, min NRMSE=0.84877

Win

argmin NRMSE=0.8, min NRMSE=0.909755

Orthogonalisation

argmin NRMSE=66, min NRMSE=0.823972

Orthonormalisation

argmin NRMSE=69, min NRMSE=0.83421

Všetko spolu

OG: argmin NRMSE=(16,29), min NRMSE=0.8187

ON: argmin NRMSE=(29,29), min NRMSE=0.823497

+ Bias

argmin NRMSE=(19,28), min NRMSE=0.838228

+Bias + I/O

 

Mackey-Glass prediction task parameter exploration

Initial network setup

Rendered by QuickLaTeX.com

Spectral radius

 

argmin NRMSE=0.875, min NRMSE=0.000247

Win

argmin NRMSE=0.1,  min NRMSE=0.00034128

Orthogonalisation

argmin NRMSE=0,  min NRMSE=0.00024545

Orthonormalisation

argmin NRMSE=10,  min NRMSE=0.00024369

Všetko spolu

 

OG: argmin NRMSE=(12,0),  min NRMSE=0.00024

ON: argmin NRMSE=(10,0),  min NRMSE=0.00024

SR/OG + bias

 

argmin NRMSE=(4,0),  min NRMSE= 0.00021

Zdrojaky:

errors.py
errors2d.py
errorsOG.py
errorsON.py
errorsWin.py

 

 

 

Mackey-Glass TE distribution

Parametre siete
Rendered by QuickLaTeX.com

TE reservoir distribution

Pozorovanie: Pri troche predstavivosti TE medzi jednotlivými neurónmi pripomína normálne rozdelenie. (teda v prípade, že TE je väčšie) A váhy rezervoárovej matice sú z normálenho rozdelenia.

TE reservoir distribution (instances of reservoir matrix)

 

TE reservoir-output distribution

 

Zdrojaky:
grafy.py
MGTEnormalized.py

 

 

16: MG, NARMA & TE cont.

To isté čo pri #20 akurát s väčším rezervoárom + pridané korelácie medzi váhami rezervoárovej matice a transfer entropie tj.  \rho(w_{ij},TE_{X_{i}\rightarrow X_{j}})

 

Parametre siete

 StandartNormalizedNormalized & OGNormalized & ON
Veľkosť rezervoiru100100100100
Train Length1000100010001000
Test length1000100010001000
Runs10101010
WN(0,1)N(0,1)N(0,1)N(0,1)
Win MGUnif[-0.5,0.5]Unif[-0.5,0.5]Unif[-0.5,0.5]Unif[-0.5,0.5]
Win NARMAUnif[0,0.4]Unif[0,0.4]Unif[0,0.4]Unif[0,0.4]
Spectral radiusunscaled0.90.90.9
Ortogonalnullnullgradient descentgradient descent
Wout trainingpinvpinvpinvpinv
Biasfalsefalsefalsefalse
Leaking rate1111
OG/ON iterations003030
OG/ON eta000.030.07*0.9**t

Parametre estimátorov

 k_historyk_taul_historyl_tauuk
TE111114
AIS11nullnullnull4

 

 

Mackey – Glass

 

 

Mean reservoir

 StandartNormalizedNormalized & OGNormalized & ON
Mackey - Glass0.00960.62450.61170.6351
NARMA0.00930.09220.08010.0945

NRMSE

 StandartNormalizedNormalized & OGNormalized & ON
Mackey - Glass1.11490.95770.67481.0472
NARMA2.41190.86640.83380.8433

InputOutput TE

 StandartNormalizedNormalized & OGNormalized & ON
Mackey - Glass-0.004-0.0210-0.0212-0.0256
NARMA-0.00130.37320.37430.3714

Mean reservoir-output TE

 StandartNormalizedNormalized & OGNormalized & ON
Mackey - Glass-0.00010.35160.36010.3532
NARMA-0.00060.00150.00140.0018

 

NARMA

  

 

AIS

 

 

Mean reservoir AIS

 StandartNormalizedNormalized & OGNormalized & ON
Mackey - Glass0.00362.16942.17632.1910
NARMA0.00230.16180.16750.1730