User's Manual
Contents
1. a run p r rnd p where rnd is a random number between 0 and 1 If negative values are al lowed the random number varies between 0 5 and 0 5 If the parameter value is 0 the parmeter is set to a random number between 0 and 1 resp 0 5 and 0 5 If the exact optimization method is selected the Variation rate increases to the Big variation rate if more than Max Steps runs without improving the best parameter set are calculated The variation is decreased after the next improvement Max runs The maximum number of runs without improvement until the optimization is reinitialized EX In this case the parameter values are reset to the values from the cev file CEvoS User s Manual September 13 2012 Chapter 1 Introduction Max steps The maximum number of runs without improvement until the variation rate is increased only EX The number of runs until the temperature resp the threshold is decreased TA SA If no improvement is reached for a specific temperature value the system is reinitialized In this case the parameter values are reset to the values from the cev file The temperature is set to its starting value Temperature threshold and annealing rate The temperature resp threshold T controls the optimization according to 5 13 resp 1 5 13 The annealing rate ra determines how fast the temperature resp threshold decreases T r T Start Rnd Starts the optimization Beginning with the param2. 4 1 Starting CEvoS in batch mode It is also possible to start CEvoS without GUI To start CEvoS in batch mode without GUI go to the directory where an existing and compiled CEMoS model the file cemos par is located and give the command cev In this case the evolution run is started To interrupt the optimization press lt Ctrl gt c The optimization run can be customized by the settings in the file cevos init see L 6 1 5 The graphical user interface GUI 1 5 1 1 5 2 1 5 3 1 5 4 Info window 1 Within this window all information during simulation etc is shown Info window 2 Within this window all parameter changes of the actual simulation are shown Best distance and Runs The actual best distance the actual absolute best distance is displayed The number of runs the number of runs after the last change of the best distance and the total number of broken runs runs which coud not be finished for any reason is displayed The best distance the difftype and the number of runs is stored in the file best dist see 2 2 Parameters Parameters IN Active Parameter Meaning from value to runs series ampli 0 000000 3 75996387236 6 0 2 390 w phase 1 0 000000 2 16117964798 0 0 1 av w regt 0 000000 080251237 f fp v Enable Series Z Pi Figure 1 2 Parameter window of CEvoS The parameters are shown in the parameter window figure 1 2 Depending on th
3. will be gradually decreased after a pre defined number of runs see 1 5 13 Simulated annealing SA The simulated annealing optimization 2001 has its roots in metal lurgy Heat causes the atoms in a metal to become unstuck from their ini tial positions a local minimum of the internal energy and wander randomly 1 5 The graphical user interface GUI CEvoS User s Manual September 13 2012 through states of higher energy Slow cooling gives them more chances of finding configurations with lower internal energy than the initial one By analogy with this physical process each step of the SA algorithm replaces the actual best run by the actual run with a probability that depends on a global parameter T called the temperature T is gradually decreased during the process annealing The best run changes almost randomly when T is large but increasingly downhill as T goes to zero The acceptance of a run which is not absolutely the actual best one takes care that local minima can be left The criterion Metropolis criterion for accepting the actual run as new best run is given by Dact lt Does T log RND where the temperature T is a positive number and RND a random number between zero and 1 To get convergence the threshold value T will be reduced after a pre defined number of runs see 1 5 13 Variation rate and big variation rate The variation rate r is the maximum variation for a positive parameter p during
4. 18459616910698 Best distance 5 836040 Actual distance 5 836040 Abs best 5 987375 Runs 20 Since last change 1 Broken runs 0 Reference file home cora MODELS TESTS cevos_distanzen3 cevos result outc Cev file home cora MODELS TESTS cevos distanzen3 cevos cevos2 cev Cin file Calculate Distances Start CEV File CIN File Reference File Start Rnd Stop Rnd Distance Pointwise 1 w g t f t 2 Positive Par iv Change all active Optimization exact threshold accepting simulated annealing Variation rate 0 1 Big Variation rate 1 0 Max runs 1000 Max steps 100 Temp Thres 1 Annealing rate 0 9 License Rename Reset to Best Reset to AbsBest Reset Tiger Graphics MoViE Exit Figure 1 1 Main window of CEvoS A single simulation with the parameterization shown in the value column can be started with Start If series is enabled a a parameter can be selected for a series run The start value the end value and the number of runs must be set Start starts the series if Enable Series is enabled All series runs will be simulated automatically The best simulation will be stored as best outc Reset to Best will set the parameters to the last best parameter set An optimization run can be started by Start Rnd The parameter values are modified by random in the given range see L 5 13 1 5 The graphical user interface GUI CEvoS User s Manual September 13 2012 1
5. Tiger Graphics e CEvos CEMoS Evolutionary System Version 4 1 User s Manual C Kohlmeier amp F Hamberg September 13 2012 CEvoS User s Manual September 13 2012 CEvoS User s Manual September 13 2012 Contents Contents 5 KSCHG 6 an 6 6 AP 9 en een 10 ERS RN EEE VEN 10 as 10 1 5 2 Info window AA 10 a nen 10 o EEE EIER 10 baa oe dpe Banca beeen anna 11 1 5 7 Calculate Distancesll 11 158 Distance 11 bad ehbngeedeeeetee 12 a 13 aos spent page 14 o oe 14 ao da eier 15 1 5 14 Stop Rnd 2 2 oo on nn 18 1 5 15 Reset to Bestll o so saors naa pur er eur e 18 1 5 16 Reset to AbsBest 2 222 2 a a a a a a a 18 154 7 Rename l e cercos se was a et 18 1 5 18 Reset s o a oo on 18 1 5 19 Tiger Graphics MoViH a se oe 34a aa 18 uo oe kia uib hia yib een ee 18 EST CEV Eiles ee a OR E 19 1522 CIN Eile sus daa aaa 19 1 5 23 Reference File 19 1 5 24 Stop Running Modell 19 1 6 The file cevos init 2 2 2 oo a a a a a a a 20 Contents CEvoS User s Manual September 13 2012 Ls das io Deia da rere 23 3 Converting field data to comparable file 25 License This program is free software you can redistribute it and or modify it under the terms of the GNU General Public License as published by the Free Soft ware Foundation either version 3 of the L
6. alculate Distances Actually six pointwise distance types are avail able Assuming that xy k 1 K i 1 N are the results from a reference file x for a variable k and yy the results from the simulation y than the distances d x y are calculated as follows 11 2 12 max Tk 1 d z y IKON 2 Wlan Yki Ok u where 77 is the time average variable k of the reference file Spitz et al 1998 2 dla EY Ey Hew Ser gt gt pe AAA AE e rr 5 ds x y ZEN EME 6 dl FX Sal FIR EN The distances are not normed to a common maximum value so that the values calculated with different distance types are not comparable Thus if a previous distance calculation is done with a different distance type a message is given and the simulation stops In this case the stored best distance and best files must be deleted before restating the simulation by Reset CEvoS User s Manual September 13 2012 Chapter 1 Introduction The number of the distance is the same as in the distance file bestdist dat 1 5 10 L2 distances The so called L2 distances work on the spectral values of the runs The trans formation of the signal into spectral values allows the comparison of to results in a manner which exceeds the pointwise method by far The comparison of simulation results and field measurements is a complicated task The normal way is to plot both within one graph and to decide with human intuition the degree of corresponde
7. arameter values of all active parameters are changed by random Otherwise only one parameter value is changed by random For every run the parameter to be changed is selected by random CEvoS User s Manual September 13 2012 Chapter 1 Introduction 1 5 13 Optimization The decision if the actual run is better than the actual best run can be done in different ways The obvious method is to compare their distances costs relative to the reference file Exact method EX The distance De is compared to the distance of the actual best run with distance Dyest The criterion for accepting the actual run as new best run is given by Dia lt Drest To avoid that the optimization stucks in a local minimum the number of runs is bounded in two ways If the number of runs without finding a new best run exceeds Max steps the variation of parameters is increased see 1 5 13 If the number of runs without finding a new best run exceeds Max runs the system is new initialized by the values in the cev file Threshold accepting TA The exact method has the disadvantage that local minima of the cost function cannot be left This can be reached if a run with a slighty worse distance is accepted Dact lt Drest 1 T where the threshold T is a positive number less than 1 Thus the actual run is accepted as new best run although its costs are higher than the cost of the actual best run To get convergence the threshold value T
8. cy The first value of a row is the lower boundary the second the upper boundary and the third the increment only needed for series runs If the original value is zero the range is set to percentage percentage If pos is set negative values are changed to zero Original negative values are set to zero if pos is set CEvoS User s Manual September 13 2012 References References Hamberg F amp Kohlmeier C 2004 CEMoS User s manual http www staff uni oldenburg de cora kohlmeier software cemoscemtk pdf Kohlmeier C amp Hamberg F 2004 MoViE User s manual http www staff uni oldenburg de cora kohlmeier software movie pdf Krug P 1992 Abstandsfunktionen in Hilbertr umen und Sch tzfunktionen in separablen Banachr umen mit Anwendungen in der mathematischen Mod ellierung Aachen Verlag Shaker Morgan B J T 2001 Applied Stochastic Modelling London Edward Arnold Spitz Y H Moisan J R Abbott M R amp Richman J G 1998 Data assimilation and a pelagic ecosystem model parameterization using time series observations Journal Of Marine Systems 16 1 2 51 68 Stearns S D 1984 Digitale Verarbeitung analoger Signale 2 edn M nchen Wien Oldenbourg
9. e file best bat is updated whenever a better distance is reached It contains the best parameterization if form of a valid batch file The parmeters are also given in cin file format best cin see Hamberg amp Kohlmeie le 2 4 The files absbest The optimization with threshold acceptance TA and simulated annealing SA leads to best results which are not absolutely the best this means that they have not the absolute minimum value for the distance but have been accepted according to the conditions of TA resp SA This is of special interest if several local minima of the cost function exist and the absolute best value and the best value are results from different local minima Thus for TA and SA the absolutely best results according to the conditions of EX are also stored in the files named absbest The fact that both are not the same is a hint that the optimization algorithm does not have converged up to know After reset the files named absbest are moved to the directory cevos_absbest in the model directory parallel to the directorycevos The files get a time stamp 2 4 The files absbest CEvoS User s Manual September 13 2012 CEvoS User s Manual September 13 2012 Chapter 3 Converting field data to comparable files 3 Converting field data to comparable files Normally field data are not available for the whole simulation period and are mostly not equidistant For the comparison with CEvoS is strictly necessary to definitio
10. e it can be defined in the sixth column a 0 1 0 5 0 05 Parameter a 0 12 In this case the simulation starts with this predefined starting value oth erwise it starts with the avearge of from and to If the randomly chosen parameter exceeds the interval the parameter is set to the predefined starting value if existing or to the average of from and to In the case of the optimization run the parameter for the increment is not evaluated but must be set for consistency 1 3 Defining a parameter set for the optimization the file cevos cev CEvoS User s Manual September 13 2012 Two auxillary scrips exist to get cev files from def file resp to define a series file with ranges sed4 CEvoS User s Manual September 13 2012 Chapter 1 Introduction 1 4 Starting CEvoS To start CEvoS go to the directory where an existing and compiled CEMoS model the file cemos par is located and give the command cev cemos par The CEvoS windows appear figure T 1 MODEL home cora MODELS TESTS cevos distanzen3 Stop Running Model Simulation time 190 000000 Simulation time 200 000000 Simulation results written to home cora MODELS TESTS cevos distanzen3 cevos resultO outc Total time O seconds Simulation time O seconds Simulation time O seconds process terminated Hx Change freq1 1 1 0156084331919653 Change ampli2 1 1 171366965875388 Change phase2 1 0 8925967658755608 Change freq2 1 0 99
11. e settings in the file cevos dat each line represents one parameter The CEvoS User s Manual September 13 2012 Chapter 1 Introduction 1 5 5 1 5 6 1 5 7 1 5 8 first column contains a checkbutton The parmeter can be enabled or disabled If the parmeter is disabled no variation of its value takes place The second column contains the name of the parameter the third its meaning as defined in the comment the column from contains the minimum value of the param eters The column value contains the actual value of the parameter The column to contains the maximum value runs and series are only valid if Enable Series is enabled Reference File A reference file must be determined for the calculation of the distances This reference file must have the structure of a CEMOS result file see B The distances are calculated for all variables which are stored in the reference file and in the simulation result files A reference file can also be made from field data For further information see Start Starts either a single simulation with the actual parameter values or a series if enabled Calculate Distances After a simulation the difference between the reference file if determined and the actual run resp runs in case of a series are printed in the info window Additionally the differences for each variable are stored in files named lt reference_name gt _ lt name gt _m lt j gt dist where lt reference_name g
12. ere are to alternatives for the syntax of this file 1 def file syntax Parameters are set similar to the settings in the def files f e CEvoS User s Manual September 13 2012 Chapter 1 Introduction a 3 0 Parameter a b 2 5 Parameter b The comments are evaluated as description In this notation only the column value in the CEvoS window is valid No series run is possible In the case of an optimization run no interval is evaluated In the case of indexed variables the values must be given separately f e ali 1 5 Parameter ali a 2 3 5 Parameter a 2 2 series syntax A range for the parameters can be given similar to the series in bat files f e a 0 1 0 5 0 05 Parameter a where the first identifier is the parameter name the second the minimum value the third the maximum value and the fourth the increment In this notation the column from contains the minimum value to the maximum value and runs the number of runs derived from the increment The actual value is set to the minimum value With these settings case series runs and optimization runs are possible If an optimization run is started the parameters are varied within the interval defined by the values in the columns from and to If from and to are equal no limitation for the parameter variation is evaluated ex ception Positive Par is enabled 1 5 11 If the optimization run should be started with a special initial valu
13. eters shown in the value column the distance to the reference file is calculated The next run is auto matically started with a modified parameter set The modifications amounts normally 10 of the absolute value of each parameter at maximum If the value in the column from differs the value given in the column to it is addi tionally proved if the parameter lies in between this interval If not the mean value of both boundaries is used as new parameter If Positive Par is en abled only positive values are used If a negative value is selected by random the old value for the parameter is fixed for the next run If the new distance is smaller than the previous the parameters are set to this better values This process is repeated until Stop Rnd is pressed The actual best parameter set is stored in bestpar bat the distance and the difftype is stored in bestdist dat the run itself is stored in best outc If the distance value could not be upgraded for ten runs modifications of 100 of the absolute value of each parameter at maximum are allowed until the distance is upgraded If the distance could not be improved for more than 10000 runs the system will be reinitialized with the original parameter values 1 5 The graphical user interface GUI CEvoS User s Manual September 13 2012 1 5 14 1 5 15 1 5 16 1 5 17 1 5 18 1 5 19 1 5 20 Stop Rnd Stops the optimization The actual run will be finished Reset to Best Se
14. icense or at your option any later version This program is distributed in the hope that it will be useful but WITHOUT ANY WARRANTY without even the implied warranty of MERCHANTA BIL ITY or FITNESS FOR A PARTICULAR PURPOSE See the GNU General Public License for more details You should have received a copy of the GNU General Public License along with this program If not see http www gnu org licenses CEvoS User s Manual September 13 2012 Chapter 1 Introduction 1 Introduction CEvoS CEMoS Evolution System is an extension of CEMoS Hamberg amp Kohlmeier which allows an optimization of model param eter values in respect to a given reference run This reference run can either be provides a preprocessing routine for transform another model run or a preprocessed run of field data The package MoViE ohne amp ber bond ing field data to the structure of a model run prepare tool data outc The basic idea is to determine the distance of a reference run and a simulation run depending on the actual parameter set and to minimize this distance by varying the parameters These distance is often also called cost function The base for the calculation of the distance are all variables which are stored in both the reference run and the simulation run The implemented calculation of the distance is described in section CEMOS allows the variation of all parameters defined in the file cevos dat which must exi
15. in best until variation is increased to bigchange only EX number of runs until temperature threshold is decreased only TA SA set maxsteps 100 number of runs without change until system is reinitialized set maxruns 1000 CEvoS User s Manual September 13 2012 Chapter 1 Introduction distance type 12dis 0 point wise 12dist 1 L2 norm for Fourier series set 12dist 0 distances cost function set difftype 1 posval 0 all values posval 1 only positive parameters set posval 1 Initializing of des simulated annealing and threshold acceptance initial temperature SA resp initial threshold TA set T 1 decreasing rate for T SA TA set rate 0 9 1 6 The file cevos init CEvoS User s Manual September 13 2012 CEvoS User s Manual September 13 2012 Chapter 2 Output Files 2 Output Files 2 1 best outc best info best log The file best outc is updated whenever a better distance is reached It con tains the results of the last best run Additionally the files best log and best info are written For details see CEMoS manual Hamberg amp Kohlmeier 2 2 best dist The file best dist is created whenever a model run is finished The first row contains the actual best distance the distance type The third row contains the number of runs The file will be deleted by Reset If no values for comparison are available the file contains dummy values 999999 2 3 best bat and best cin Th
16. n a reference file which has the same structure as the result files itself The easiest way to do this is to create the reference file by the model itself In the case of field data this is normally not useful The TigerGraphics tool MoViE 2004 provides the possibility to convert field data to the format needed for CEvoS see MoViE Prepare Tool data outc The values are linearly interpolated The startime endtime outdelt and cycle must be set to the values of the model for which the reference is designed If the interval of the resultfile exceeds the field data the values for the missing time stamps at the beginning and at the end are set to 9999 0 All values which are exactly 9999 0 will not be considered during the comparison If the comparison should be done with L2 distances the linear interpolated reference file is automatically converted to spectral values WARNING Actually it is not possible to select a _dpk outc file for refer ence see MoViE Prepare Tool gt data Fourier Chapter 3 Converting field data to comparable files CEvoS User s Manual September 13 2012 CEvoS User s Manual September 13 2012 Chapter 4 Auxillary scripts 4 Auxillary scripts 4 1 def2cev The script def2cev converts CEMoS definition files extension def into CEvoS parmeter files extension cev All lines not needed by CEvoS will be deleted Thus only the statments real_par and real_ind_par are evaluated Li
17. nce This is of course a very subjective decision because from psychology it is well known that people tend to see what they expect or like to see Such it is necessary to give an assistance to the human intuition for the ex ploitation of the characteristic differences and correspondings of the theoretical results and the real life This is only possible if not only single points are compared but the time series are compared at a whole Only then f e phase shifts can be taken into account The L2 distances are calculated as follows In a first step a fast Fourier analyses of the reference file and the simulation run is made resulting in _pk outc file containing the spectral values for every stored variable This method works only successfully if the data are periodic in time and the stored time interval covers the period The maximum number of heer plex Fourier coefficients k is concerning the sampling theorem Steam 1984 1984 half the number of samples Assuming N samples the Fourier coefficients are given by a vector g go 91 Jak R with k 3 1 Thus the original value of a variable x at time t is given by EE 9211 cos lt ga sin It a t Tor m gt The norm of a time series x ae the Fourier coefficients fo fi fox which is an element of the room L2 hence the name L2 distances is given by Ill 4 Yi and the scalar ze of two time series x and y with the Fourier coefficient
18. nes defining real_ind_par are splitted into single lines f e def2cev xyz def converts the file xyz def hreal_ind_par ampli 3 1 2 3 Amplitude phase 3 0 0 25 0 5 Phase freq 3 1 1 1 Frequency globals sinus 3 change phase 1 3 10 0 5 0 5 to Xyz cev ampli 1 1 Amplitude ampli 2 2 Amplitude ampli 3 3 Amplituder phase 1 0 Phase phase 2 0 25 Phase phase 3 0 5 Phase freq 1 1 Frequency freq 2 1 Frequency freq 3 1 Frequency It is strictly recommended to manually correct the automatically converted files afterwards 4 2 cev2ser 4 2 cev2ser CEvoS User s Manual September 13 2012 The script cev2ser converts a regular cev file in parameter notation see 1 3 to a cev file in series notation The script requires up to 3 arguments First the filename to be converted second the percentage for the modification of the parameter and third optional the flag pos if only positive including zero values are allowed The file xyz cev is converted by cev2ser xyz cev 10 pos to the file xyz_ser cev ampli 1 0 ampli 2 1 ampli 3 2 0 0 0 phase 1 phase 2 phase 3 9 1 1 0 1 Amplitude 8 2 2 0 2 Amplitude 7 3 3 0 3 Amplitude 0 1 0 1 Phase 225 0 275 0 025 Phase 45 0 55 0 05 Phase freq 1 0 9 1 1 0 1 Frequency freq 2 0 9 1 1 0 1 Frequency freq 3 0 9 1 1 0 1 Frequen
19. ntrolling the simulation can be defined see section 19 7 in Hamberg amp Kohlmeie 2004 for details If cancel is pressed no cin file is evaluated Reference File A outc file can be defined This is taken as the goal for the training of the model under regard All distances pointwise or L2 are computed against the values held in this file see B If cancel is pressed no reference file is selected and the calculation of distances will not work Stop Running Model A running simulation can be stopped If the random process is running CEvoS continues with the next parameter set 1 6 The file cevos init CEvoS User s Manual September 13 2012 1 6 The file cevos init The file lt model_dir gt cevos cevos init is optional If the file exist the default settings are overruled The file can contain the settings for the control of the optimization runs It must be written in Tcl TK notation Example cevos file with absolute path Default set cev_file cev_resultdir cevos cev set cev_file cev_resultdir cevos cev cin file with absolute path Default none set cin_file Reference file with absolute path Default none set cev_reference_file 0ptimization SA 0 exact EX SA 1 threshold acceptance TA SA 2 simulated annealing SA set SA 0 Variation rate of Parameters set smallchange 0 1 big variation of parameters only SA 0 set bigchange 1 0 number of runs without change
20. s fo fi Jar resp go fi 92k 18 given by 2k lt y gt fogo 5 figi cos y p 1 0 1 5 The graphical user interface GUI CEvoS User s Manual September 13 2012 with y arctan 2 and y arctan ga ji 2l 921 1 The following distance functions are implemented 1 di f 9 f gll IF gll 2 da f 9 F gll min I fil Ilall 3 ds f g F gll Olgll II IF all 4 da f g 1 lt f g gt IIFI Ilall 5 ds f 9 FI Ilall lt Hs gt tl Al lgll lt g gt 6 de f g arccos lt f g gt IFII all 7 d f 9 MAN all 8 de f g max HF1 Mol Mol ALFI 1 9 do f g Aide f g A2d7 f 9 MM A2 1 The distances d d3 dy and dg focus on differences in the amplitudes while distances d4 dg measure the phase shift All distances are invariant concern ing rotation and stretching some are definite surjective or fulfill the triangle inequality Only dy fulfills all these properties 1999 has shown that dy is nearest to the human estimation WARNING The preparation for the calculation of L2 distances matrix in version can take a very long time after the first model run The inverted matrix is stored so that the following calculations are quite faster 1 5 11 Positive Par If this option is set the parameter variations allows only non negative values 1 5 12 Change all active If this option is set the p
21. ser s Manual September 13 2012 The simulation of a model is similar to the simulation within the CEMTK shell All parameter changes are read from the cemos par resp def files Al ternatively can the simulation be started using a cin file In both cases the values of the parameters defined in cevos cev are overruled by the values defined in cevos cev 1 1 The idea of parameter adaption The optimization of parameter can be summarited as follows 1 Defining the parameters to be optimized see 1 3 2 Defining a target a reference run This reference run can either be another model run or a preprocessed run of field data see 3 Determing a cost function or distance function see 1 5 8 4 Determing an optimization method and escape criteria see 1 5 13 1 2 Requirements and Installation of CEvoS CEvoS is part of TigerGraphics TigerPack and an addendum of CE MoS and MoViE For installtion see CEMoS manual After the installation of CEMoS CEMTK and MoViE CEvoS is available and ready for use Before starting the parameters to be varied have to be defined 1 3 Defining a parameter set for the optimization the file cevos cev Before starting CEvoS the directory lt model_dir gt cevos must exist containin at least on file with the extension cev This file specifies the model param eters which should be trained If the file is named cevos cev it is read by default Files with other names can be selected by CEV Files Th
22. st defined in a directory lt model_dir gt cevos This parameters are either varied by a defined series or by random After every calculation the resulting distance is compared to the former distance The best parameterization is stored in bestpar bat the best distance in bestdist dat and the run itself in best outc For the decision which is the best distance three optimization algorithms are imple mented section 1 5 13 If the system has been started by Start Rnd section one parameter defined in cevos dat is varied randomly by 10 of the value at maximum The optimization cycle will run until Stop Rnd is pressed If the system is restarted again the optimization process will start with the last best parameterization CEvoS also allows sensitivity analyses For this a simulation series can be started where a selected parameter is varied by a pre detrmined increment in apre determined range If a series has been calculated CEvoS stops automat ically after this series The distances to the reference run are stored All simulation runs are stored in lt model_dir gt cevos The last run is always named result0 outc If it is the best run it is copied to best outc In case of a random start only the last run is kept to avoid storage overflow WARNING Other files than files used by CEvoS in the directory cevos are deleted during the run of CEvoS Do not store any infor mation in the directory cevos 1 1 The idea of parameter adaption CEvoS U
23. t is the name of the selected reference fileand lt name gt is the name of the result file and lt j gt the index of the distance Distances Two types of distances are available pointwise distances and L2 distances The pointwise distances compare the reference file and the actual simulation at every stored time step The L2 distances make Fourier analyses of both the reference file and the actual run and compare the spectral values It is either a pointwise or a L2 distance active The active distances is displayed The distances are not normed to a common maximum value so that the values calculated with different distance types are not comparable Thus if a previous 1 5 The graphical user interface GUI CEvoS User s Manual September 13 2012 distance calculation is done with a different distance type a message is given and the simulation stops In this case the stored best distance and best files must be deleted before restating the simulation by Reset 1 5 9 Pointwise distances The pointwise distance of the reference file and the simulation run is deter mined in two steps First the distance between every single stored vari able is determined for the stored time window and normed by the num ber of time steps Afterwards the resulting distance values for the differ ent variables are averaged The resulting number is the distance of the two files The intermediate distances for the variables are stored in the file by C
24. ts the parameters and the distance to the last best values if exist Reads the distance and the counter from file Reset to AbsBest Sets the parameters and the distance to the absolute best values if exist Reads the distance and the counter from file The absolue best values and the best values differ if simulated annealing or threshold accepting is active Rename All files belonging to a run can be renamed to keep it The renamed files will be stored in the directory lt model_dir gt cevruns Reset Deletes all best files except the file bestpar bat in the directory cevos After reset the files named absbest are moved to the directory cevos_absbest in the model directory parallel to the directorycevos The files get a time stamp A reset should be done before changing the distance type L 5 8 Tiger Graphics MoViE Starts the TigerGraphics tool MoViE and opens directly the compare tool for the reference file and the last best run For further information see Kohlmeier amp Hamberg A Exit Closes the CEvoS shell The best files are still available and can be read after restarting CEvoS CEvoS User s Manual September 13 2012 Chapter 1 Introduction 1 5 21 1 5 22 1 5 23 1 5 24 CEV File A cev file determing the parameters can be defined By default the file cevos cev is read After changing the file the parameter window will be new initialized CIN File A cin file co
Download Pdf Manuals
Related Search