Which algorithm can I use in Weka to find best parameters that will maximize the output?

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Which algorithm can I use in Weka to find best parameters that will maximize the output?

Ayşe Arslan
Hello,

I am working on parallel I/O operations, trying to improve I/O performance which is measured with obtained bandwidth. In my project, there are different configuration parameters, my aim is to find configuration set that will give best "bandwidth" value.

My 4 configuration parameters can take different values as follows:
parameter1 : automatic, enable, disable
parameter2 : automatic, enable, disable
parameter3 : 1, 2, 4, 6, 8, 10, 12, 16
parameter4 : 1048576, 2097152, 4194304, 8388608, 16777216, 33554432

I have 54 different data size.

So I want to measure bandwidth for all these combinations for all data sizes. This is computationally expensive function, but I do. Then for each file type (depending on data size) I extract configuration set that gives best bandwidth.

Finally I get such a log after execution for each combination, a long log file includes all configuration possibilities and obtained bandwidth with that configuration.

{"file_type":38","operation_type":"collective","bytes":120000000,"duration":0.240724,"bandwidth":475.402854,"conf_set":[{"parameter1":"automatic"},{"parameter2":"automatic"},{"parameter3":"1"},{"parameter4":"2097152"}]

In Weka, how can I work on the data? I have no class attribute. I am trying to  find parameter set that give best bandwith value. Any ideas?

Thank you,
Ayse

_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to: To unsubscribe send an email to [hidden email]
To subscribe, unsubscribe, etc., visit
https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|

Re: Which algorithm can I use in Weka to find best parameters that will maximize the output?

Eibe Frank
I'm not entirely sure I understand what you are trying to do, but people often build a surrogate function to find good parameters for a function that is expensive to evaluate. In your case, based on the data you have collected, you may be able to build a regression model using WEKA that uses the four configuration parameters plus the other attributes (file size and type?) as independent variables to estimate bandwidth as the dependent variable. Once you have that regression model (i.e., the surrogate function), you can then apply a black-box optimization method such as a genetic algorithm to optimise this function as desired. This latter optimization process would have to be implemented from scratch or performed using some suitable optimization library. It cannot be performed directly in WEKA.

Cheers,
Eibe

On Tue, Dec 3, 2019 at 8:15 AM Ayşe Arslan <[hidden email]> wrote:
Hello,

I am working on parallel I/O operations, trying to improve I/O performance which is measured with obtained bandwidth. In my project, there are different configuration parameters, my aim is to find configuration set that will give best "bandwidth" value.

My 4 configuration parameters can take different values as follows:
parameter1 : automatic, enable, disable
parameter2 : automatic, enable, disable
parameter3 : 1, 2, 4, 6, 8, 10, 12, 16
parameter4 : 1048576, 2097152, 4194304, 8388608, 16777216, 33554432

I have 54 different data size.

So I want to measure bandwidth for all these combinations for all data sizes. This is computationally expensive function, but I do. Then for each file type (depending on data size) I extract configuration set that gives best bandwidth.

Finally I get such a log after execution for each combination, a long log file includes all configuration possibilities and obtained bandwidth with that configuration.

{"file_type":38","operation_type":"collective","bytes":120000000,"duration":0.240724,"bandwidth":475.402854,"conf_set":[{"parameter1":"automatic"},{"parameter2":"automatic"},{"parameter3":"1"},{"parameter4":"2097152"}]

In Weka, how can I work on the data? I have no class attribute. I am trying to  find parameter set that give best bandwith value. Any ideas?

Thank you,
Ayse
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to: To unsubscribe send an email to [hidden email]
To subscribe, unsubscribe, etc., visit
https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to: To unsubscribe send an email to [hidden email]
To subscribe, unsubscribe, etc., visit
https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html