Genetic Programming

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Genetic Programming

salima
This post has NOT been accepted by the mailing list yet.
Dear Weka friends,


I am focusing on Genetic programming modelling approach using WEKA.
I am doing a research about the classification of customer credit using the german credit data set.
 Therefore I made a genetic programming, where  WEKA gives me weight. Could you please help me to interpret them in the right and proper manner? I shall be very glad if you help me with this problem. MANY THANKS in ADVANCE I do not find any experts of WEKA to consults.

I need shows the weight of each characteristics of the data set in order to calculate the credit scoring function.
 
Looking forward to your explanation.

Regards,
salima


Thanks in advance=== Run information ===

Scheme:       weka.classifiers.geneticprogramming.GeneticProgramming
Relation:     german_credit-weka.filters.supervised.attribute.Discretize-Rfirst-last
Instances:    1000
Attributes:   21
              checking_status
              duration
              credit_history
              purpose
              credit_amount
              savings_status
              employment
              installment_commitment
              personal_status
              other_parties
              residence_since
              property_magnitude
              age
              other_payment_plans
              housing
              existing_credits
              job
              num_dependents
              own_telephone
              foreign_worker
              class
Test mode:    10-fold cross-validation

=== Classifier model (full training set) ===


Details of Genetic Programming Experiment

Database name : german_credit-weka.filters.supervised.attribute.Discretize-Rfirst-last
Number of attributes : 21
Number of instances for training: 500
Number of instances for validation: 500

 .: Genetic Parameters :.

1. Statistical Normalization Pre-processor : normalizes data to have mean = 0.0 and standard deviation = 1.0
2. Population size : 100
3. Proportion of training data kept for validation : 0.5
4. Program rules :
Maximum depth of program tree: 5
Number of Inputs for programs: 20
Number of available classes: 2
Number of possible content (input and functions) for nodes: 34
Proportion of input in terminals : 0.5
Proportion of float constants (comprared to integer constants): 1.0
ADFs are not used.
Arguments types : Arithmetic, Test, Any,
Function Table : [+, -, /, *, If, >, <, Pow, &, |, Max, Min, Exp, Log]
5. One Class Weight Evaluator : returns confidence (0.0 to 1.0) of instance being of current class
6. Tree Population (ramped) Half and Half Initializer : initializes a population of program trees using the Full and Grow method, creating trees with depth from 2 to maximum depth, with an equal number of programs for each depth.
Type of programs : MainProgramTree
7. Fitness-proportionnal Selector : selects programs with probability based on fitness (with fitness / sumFitness probability for each program).
8. Keep Bests Elite Manager : keeps 1 best programs since beginning of run in memory.
9. Select After Evolution Controller : we create a new population of 100 children programs. From the total of parents and the children is selected enough individual to produce the next generation of programs.
10. Tree Crossover Operator : "crosses" two programs together, switching a random sub-tree from each of the programs. Results are two offsprings.
        Proportion = 0.9
        Number of parents = 2
        Number of children = 2
11. Tree Mutation Operator : mutates a whole randomly selected sub-tree from a program.
        Proportion = 0.07
        Number of parents = 1
        Number of children = 1
12. New Program Tree Operator : creates completely new programs.
        Proportion = 0.03
        Number of parents = 1
        Number of children = 1

Target fitness = 0.9
Maximum time = 0.033 minutes
Max generations = 20

Genetic run finished.
Creation of a multiclass classifier from multiple one-class classifier (multiple GP runs).

*** CLASS good ***

Evolution ended at 20th generation. (T = 0.009666666666666667 minutes, F = 0.7)
Elite program no.0, size = 1, training fitness = 0.712, validation fitness = 0.688
Error = 0.30000000000000004
Co 2,301

Evolution ended at 20th generation. (T = 0.02285 minutes, F = 0.6836034545898437)
Elite program no.1, size = 7, training fitness = 0.6864161529541015, validation fitness = 0.6807907562255859
Error = 0.363
Exp( *( Exp( checking_status ), Pow( checking_status, checking_status ) ) )

Evolution ended at 20th generation. (T = 0.026033333333333332 minutes, F = 0.6185776901245117)
Elite program no.2, size = 13, training fitness = 0.5969524536132812, validation fitness = 0.6402029266357422
Error = 0.30300000000000005
Exp( Pow( *( checking_status, Max( duration, checking_status ) ), Pow( +( checking_status, employment ), Log( duration ) ) ) )

Evolution ended at 20th generation. (T = 0.02755 minutes, F = 0.5937481079101562)
Elite program no.3, size = 8, training fitness = 0.6156075363159179, validation fitness = 0.5718886795043945
Error = 0.42600000000000005
Exp( Pow( +( credit_history, duration ), Exp( Log( employment ) ) ) )

Evolution ended after 0.03406666666666667 minutes. (G = 19, F = 0.6193760986328125)
Elite program no.4, size = 9, training fitness = 0.6277621383666993, validation fitness = 0.6109900588989258
Error = 0.40700000000000003
Exp( Pow( +( credit_history, savings_status ), Exp( Pow( other_payment_plans, other_payment_plans ) ) ) )


*** CLASS bad ***

Evolution ended at 20th generation. (T = 0.005583333333333333 minutes, F = 0.7)
Elite program no.0, size = 1, training fitness = 0.712, validation fitness = 0.688
Error = 0.30000000000000004
num_dependents

Evolution ended at 20th generation. (T = 0.012866666666666667 minutes, F = 0.6836796417236328)
Elite program no.1, size = 15, training fitness = 0.6931598663330079, validation fitness = 0.6741994171142578
Error = 0.355
|( |( <( Min( Co 0,658, other_payment_plans ), Co 0 ), If( >( other_payment_plans, checking_status ), >( property_magnitude, checking_status ) ) ), Co 0 )

Evolution ended at 20th generation. (T = 0.03025 minutes, F = 0.629619831085205)
Elite program no.2, size = 15, training fitness = 0.6194626998901367, validation fitness = 0.6397769622802735
Error = 0.31000000000000005
|( <( Min( Co 0,658, checking_status ), Co -0,903 ), If( >( Min( credit_amount, job ), checking_status ), -( savings_status, credit_amount ) ) )

Evolution ended at 20th generation. (T = 0.0316 minutes, F = 0.5687900276184081)
Elite program no.3, size = 15, training fitness = 0.5971943817138672, validation fitness = 0.5403856735229492
Error = 0.406
|( |( <( Min( checking_status, checking_status ), Log( housing ) ), Co 0 ), <( Log( /( checking_status, credit_amount ) ), Co 0 ) )

Evolution ended after 0.0345 minutes. (G = 20, F = 0.5805260124206544)
Elite program no.4, size = 16, training fitness = 0.5860976715087891, validation fitness = 0.5749543533325195
Error = 0.481
|( |( <( +( savings_status, housing ), /( housing, savings_status ) ), Co 0 ), <( Log( /( checking_status, Co 0,658 ) ), Co 0 ) )


Weights from orchestration of programs

         Program weights for class 0 : PW0-0=0.8472978603872034, PW0-1=0.5623668213073126, PW0-2=0.8330526052511545, PW0-3=0.29819005005019583, PW0-4=0.3763812135551304
         Program weights for class 1 : PW1-0=0.8472978603872034, PW1-1=0.5971325273203567, PW1-2=0.8001193001121129, PW1-3=0.38052615976091214, PW1-4=0.07603661306012757

Time taken to build model: 15 seconds



Loading...