Auto-weka problem

classic Classic list List threaded Threaded
21 messages Options
12
Reply | Threaded
Open this post in threaded view
|

Auto-weka problem

Nilhee2000
Hi guys,

I have installed Auto-weka from packet manager. I have used auto-weka and it
provided me the results. However, when I wanted to save the model provided
by the auto-weka, the following message was shown:

"java.io.notserializableexception: autoweka.configuration collection"

Could anyone help how could I solve this problem in order to be able to save
my model?

Thanks,



--
Sent from: https://weka.8497.n7.nabble.com/
_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|

Re: Auto-weka problem

Eibe Frank-2
Administrator

Based on the error message, it looks like one of the Auto-WEKA classes does not implement the Serializable interface.  You might have to rebuild the classifier outside Auto-WEKA based on the configuration output by Auto-WEKA.

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Saturday, 17 August 2019 12:29 PM
To: [hidden email]
Subject: [Wekalist] Auto-weka problem

 

Hi guys,

 

I have installed Auto-weka from packet manager. I have used auto-weka and it

provided me the results. However, when I wanted to save the model provided

by the auto-weka, the following message was shown:

 

"java.io.notserializableexception: autoweka.configuration collection"

 

Could anyone help how could I solve this problem in order to be able to save

my model?

 

Thanks,

 

 

 

--

Sent from: https://weka.8497.n7.nabble.com/

_______________________________________________

Wekalist mailing list

Send posts to: [hidden email]

To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/mailman/listinfo/wekalist

List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

 


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

Re: Auto-weka problem

Tony D
In reply to this post by Nilhee2000
Greetings!
I have the same issue. My installation seems to let me save the model from
auto weka but refuses to load it for later use. Are there any updates? If I
must use the results buffer to recreate the classifier, are all the bits
there to do so without guessing? Is there an add-on package that will allow
reading the model so as to get something from there?
Many thanks!

Tony



--
Sent from: https://weka.8497.n7.nabble.com/
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Michael Hall


> Are there any updates?

Java 9+ reflection warning omitted

java -cp /Applications/weka-3-9-5/weka.jar weka.core.WekaPackageManager -package-info repository Auto-WEKA | grep Version:
    Version: 2.6.1

java -cp /Applications/weka-3-9-5/weka.jar weka.core.WekaPackageManager -package-info installed Auto-WEKA | grep Version:
java.lang.NullPointerException: Cannot invoke "weka.core.packageManagement.Package.getPackageMetaData()" because the return value of "weka.core.WekaPackageManager.getInstalledPackageInfo(String)" is null
        weka.core.WekaPackageManager.printInstalledPackageInfo(WekaPackageManager.java:2505)

Would show you your installed version. I don’t have one. You can then compare the versions to see if there has been an update.

I added a command to my HalfPipe application to automate this checking for all my installed packages indicating any updates…

weka.pkgmgr.check
WekaPackageCheck: mismatch 1.6.0   1.7.1 wekaDeeplearning4j Weka wrappers for Deeplearning4j
WekaPackageCheck: mismatch 1.2.30   1.3.30 RPlugin Execute R scripts and learning algorithms
WekaPackageCheck: mismatch 1.0.12   1.0.13 wekaPython Provides integration with CPython in Weka.
WekaPackageCheck: mismatch 2019.10.4   2020.2.17 multisearch MultiSearch Parameter Optimization
WekaPackageCheck: done


_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Michael Hall


On Feb 27, 2021, at 7:58 PM, Michael Hall <[hidden email]> wrote:



Are there any updates?


Forgetting the obvious but the Weka package manager gui also has the prior information.

I don’t think Weka packages can have add-ons. I believe they can have some kind of dependencies, not sure what they are.

I wouldn’t guess on serialization errors without complete stack traces. If that gives you a clue to the class not being serialized.

I don’t see any serialization type issues off hand.
There doesn’t appear to have been recent activity.


_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Tony D
"I wouldn’t guess on serialization errors without complete stack traces. If
that gives you a clue to the class not being serialized."

Um... how do I extract that trace? I am ~not~ a programming type - not for a
while (I mostly checked out back at FortranIV and Pascal)... Happy to go get
that trace if it will help.

Also - the output of the autoweka routine DID write a model file - opens in
wordpad showing a mache of readable and not ASCII. I put that where I could
find it - should it be somewhere specific? I am getting the idea about paths
in java...

T






--
Sent from: https://weka.8497.n7.nabble.com/
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Michael Hall


> On Mar 1, 2021, at 2:26 PM, Tony D <[hidden email]> wrote:
>
> "I wouldn’t guess on serialization errors without complete stack traces. If
> that gives you a clue to the class not being serialized."
>
> Um... how do I extract that trace? I am ~not~ a programming type - not for a
> while (I mostly checked out back at FortranIV and Pascal)... Happy to go get
> that trace if it will help.

I would probably check weka.log first.
Should be at…
<your user directory>/wekafiles/weka.log
See if there are any stack traces showing there.

>
> Also - the output of the autoweka routine DID write a model file - opens in
> wordpad showing a mache of readable and not ASCII. I put that where I could
> find it - should it be somewhere specific? I am getting the idea about paths
> in java…

I thought we were talking about a problem saving a serialized model. If you have one you should be good?
I don’t usually work with models. However you usually open them should work?


_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Tony D
Michael Hall wrote
>> On Mar 1, 2021, at 2:26 PM, Tony D &lt;

> tdecarmine@

> &gt; wrote:
>>
>> "I wouldn’t guess on serialization errors without complete stack traces.
>> If
>> that gives you a clue to the class not being serialized."
>
>> Happy to go get
>> that trace if it will help.
>
> I would probably check weka.log first.
> Should be at…
> <your user directory>
> /wekafiles/weka.log
> See if there are any stack traces showing there.
>
>>
>> Also - the output of the autoweka routine DID write a model file - opens
>> in
>> wordpad showing a mache of readable and not ASCII. I put that where I
>> could
>> find it - should it be somewhere specific? I am getting the idea about
>> paths
>> in java…
>
> I thought we were talking about a problem saving a serialized model. If
> you have one you should be good?
> I don’t usually work with models. However you usually open them should
> work?

I have some 400kb of log from the last run. I will look for any serialize
entries.

However, I suspect the model write operation isn't there as I saved it
manually afterwards with the usual right click in the results window. No
error then but i get the NotSerializable error shown elsewhere when i try to
load it into CLassifier again with the usual right click.

I'd held hope that maybe it could be seen from the classifier without saving
and recalling right after the autoweka run, but no.

I also tried to reconstitute the classifier with the visible parameters from
the results buffer but no luck there.  is there a way to extract a saved
model to human readable form?




--
Sent from: https://weka.8497.n7.nabble.com/
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Michael Hall


On Mar 2, 2021, at 10:55 AM, Tony D <[hidden email]> wrote:

Michael Hall wrote
On Mar 1, 2021, at 2:26 PM, Tony D &lt;

tdecarmine@

&gt; wrote:

"I wouldn’t guess on serialization errors without complete stack traces.
If
that gives you a clue to the class not being serialized."

Happy to go get
that trace if it will help.

I would probably check weka.log first.
Should be at…
<your user directory>
/wekafiles/weka.log
See if there are any stack traces showing there.


Also - the output of the autoweka routine DID write a model file - opens
in
wordpad showing a mache of readable and not ASCII. I put that where I
could
find it - should it be somewhere specific? I am getting the idea about
paths
in java…

I thought we were talking about a problem saving a serialized model. If
you have one you should be good?
I don’t usually work with models. However you usually open them should
work?

I have some 400kb of log from the last run. I will look for any serialize
entries.

However, I suspect the model write operation isn't there as I saved it
manually afterwards with the usual right click in the results window. No
error then but i get the NotSerializable error shown elsewhere when i try to
load it into CLassifier again with the usual right click. 

I'd held hope that maybe it could be seen from the classifier without saving
and recalling right after the autoweka run, but no. 

I also tried to reconstitute the classifier with the visible parameters from
the results buffer but no luck there.  is there a way to extract a saved
model to human readable form? 


I would of thought there would be an error on the save but it’s been a while since I did anything with Serialization.

I did do this sometime back which might give you a stack trace if run command line.


It was meant to recreate models after a version change broke some serialization.

So weka.jar would need to be in class path and then

-n would also have to indicate where weka.jar is, since you don’t have a new version jar to point to in this case
-t would be a training set to create a new model with the new jar on.
-d would be where to save the new model (destination?)
The final argument would be the existing model file.

So something like…

java -cp .:weka.jar WekaSerialConverter -n weka.jar -d newModel.m -t training.arff oldModel.m

It should read the old model, try to figure out enough to re-run the classifier, and then run the classifier to recreate a new model with the new jar. 
If there’s a serialization error involved here I think it should throw an exception somewhere.

 


_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Tony D
Michael Hall wrote
>> On Mar 2, 2021, at 10:55 AM, Tony D &lt;

> tdecarmine@

> &gt; wrote:
>> snip <
>
> I would of thought there would be an error on the save but it’s been a
> while since I did anything with Serialization.
>
> I did do this sometime back which might give you a stack trace if run
> command line.
>
> http://mikehall.pairserver.com/WekaSerialConverter.java
> &lt;http://mikehall.pairserver.com/WekaSerialConverter.java&gt;
>
> It was meant to recreate models after a version change broke some
> serialization.
>
> So weka.jar would need to be in class path and then
>
> -n would also have to indicate where weka.jar is, since you don’t have a
> new version jar to point to in this case
> -t would be a training set to create a new model with the new jar on.
> -d would be where to save the new model (destination?)
> The final argument would be the existing model file.
>
> So something like…
>
> java -cp .:weka.jar WekaSerialConverter -n weka.jar -d newModel.m -t
> training.arff oldModel.m
>
> It should read the old model, try to figure out enough to re-run the
> classifier, and then run the classifier to recreate a new model with the
> new jar.
> If there’s a serialization error involved here I think it should throw an
> exception somewhere.

So, me being a JavaNoob, I have placed copies of the WekaSerialConverter
everywhere I think might be in a path... no luck. Java can't find it. tried
to run from several directories - also no luck. I realize thisis a Java
specific thing... Feel free to point me at a ref guide (RTFM, Noob!) for
this part of the operation.  

FYI - I also tried fracpete model viewer installed as a tab in Weka -
nothing happens when i try to open the model. no error - no nothing.
:(




--
Sent from: https://weka.8497.n7.nabble.com/
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Michael Hall


On Mar 6, 2021, at 4:09 PM, Tony D <[hidden email]> wrote:

Michael Hall wrote
On Mar 2, 2021, at 10:55 AM, Tony D &lt;


-n would also have to indicate where weka.jar is, since you don’t have a
new version jar to point to in this case
-t would be a training set to create a new model with the new jar on.
-d would be where to save the new model (destination?)
The final argument would be the existing model file.

So something like…

java -cp .:weka.jar WekaSerialConverter -n weka.jar -d newModel.m -t
training.arff oldModel.m



So, me being a JavaNoob, I have placed copies of the WekaSerialConverter
everywhere I think might be in a path... no luck. Java can't find it. tried
to run from several directories - also no luck. I realize thisis a Java
specific thing... Feel free to point me at a ref guide (RTFM, Noob!) for
this part of the operation.  

What I pointed you to was java source you would need to compile it to a class first with something like…
javac -cp /where-ever/is/your/weka.jar WekaSerialConverter.java

Then run it with something like...
java -cp .:/where-ever/is/your/weka.jar WekaSerialConverter -n /where-ever/is/your/weka.jar -d /where-ever/you/want/newModel.model -t /where-ever/is/your/training.arff /where-ever/is/your/oldModel.model

You should only need it one place but the paths to the files need to be right. If you are not familiar with command line or java this might not be something for you to try. 


FYI - I also tried fracpete model viewer installed as a tab in Weka -
nothing happens when i try to open the model. no error - no nothing.
:(


Again, errors probably would be in the weka.log. Unless the application catches them and tells you. 
Searching weka.log for ‘exception' maybe if it’s too big to spot the stack traces, if there is even a stack trace, I guess no guarantee.
But if the stack trace is there it would give us something to go on.

Or if you can attach or put the model file somewhere we could get at?

_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Tony D
Michael Hall wrote
>> On Mar 6, 2021, at 4:09 PM, Tony D &lt;

> tdecarmine@

> &gt; wrote:
>>
>>snip<
>
>
> What I pointed you to was java source you would need to compile it to a
> class first with something like…
> javac -cp /where-ever/is/your/weka.jar WekaSerialConverter.java
>
> Then run it with something like...
> java -cp .:/where-ever/is/your/weka.jar WekaSerialConverter -n
> /where-ever/is/your/weka.jar -d /where-ever/you/want/newModel.model -t
> /where-ever &lt;http://where-ever.is/&gt;/is/your/training.arff
> /where-ever/is/your/oldModel.model
>
> You should only need it one place but the paths to the files need to be
> right. If you are not familiar with command line or java this might not be
> something for you to try.
>
>>
>> FYI - I also tried fracpete model viewer installed as a tab in Weka -
>> nothing happens when i try to open the model. no error - no nothing.
>> :(
>>
>
> Again, errors probably would be in the weka.log. Unless the application
> catches them and tells you.
> Searching weka.log for ‘exception' maybe if it’s too big to spot the stack
> traces, if there is even a stack trace, I guess no guarantee.
> But if the stack trace is there it would give us something to go on.
>
> Or if you can attach or put the model file somewhere we could get at?

OK - I have the JDK loaded and fiddling with compiling the
wekaserialconverter. throws some 7 errors, mostly Not Found errors fixed by
the compiler getting the files - but at least one is killing the process. I
will look into that more shortly.

My model file here:
testseta.model <https://weka.8497.n7.nabble.com/file/t7321/testseta.model>  
or so I hope.
curious about forensics to be gathered...

I know the etiquette says not to give thanks, but, THANKS!



--
Sent from: https://weka.8497.n7.nabble.com/
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Michael Hall


My model file here:
testseta.model <https://weka.8497.n7.nabble.com/file/t7321/testseta.model>  
or so I hope.
curious about forensics to be gathered...

I know the etiquette says not to give thanks, but, THANKS!

For WekaSerialConverter to run as-is it would also need a training file compatible with the one that created the model file.
I can do some modifications so it just tries to read the model file.

_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Michael Hall


On Mar 11, 2021, at 4:50 PM, Michael Hall <[hidden email]> wrote:



My model file here:
testseta.model <https://weka.8497.n7.nabble.com/file/t7321/testseta.model>  
or so I hope.
curious about forensics to be gathered...

I know the etiquette says not to give thanks, but, THANKS!

For WekaSerialConverter to run as-is it would also need a training file compatible with the one that created the model file.
I can do some modifications so it just tries to read the model file.

Forgot I had an option to just read a model file if that was the only argument. Been a while.

java.io.WriteAbortedException: writing aborted; java.io.NotSerializableException: autoweka.ConfigurationCollection

Which if I remember right was the original error reported. When I did some quick checking then it didn’t seem to be a autoweka class. But it apparently got serialized into the model somehow, but doesn’t deserialize. I might look a little more.

java.io.WriteAbortedException: writing aborted; java.io.NotSerializableException: autoweka.ConfigurationCollection
at java.base/java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1719)
at java.base/java.io.ObjectInputStream.defaultReadFields(ObjectInputStream.java:2540)
at java.base/java.io.ObjectInputStream.readSerialData(ObjectInputStream.java:2390)

It is still sort of strange to me how a write gets involved here. 
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Michael Hall

Which if I remember right was the original error reported. When I did some quick checking then it didn’t seem to be a autoweka class. But it apparently got serialized into the model somehow, but doesn’t deserialize. I might look a little more.


I’m not sure what I searched last time but this time googled


It does turn up on automl. I think someone associated with that had responded earlier as well?

public class ConfigurationCollection
extends XmlSerializable

It appears to implement it’s own XML flavored serialization? 

You might check with them. Maybe raise on issue on their GitHub. If the earlier atonal poster isn’t currently following this.



_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Michael Hall


On Mar 11, 2021, at 5:15 PM, Michael Hall <[hidden email]> wrote:


Which if I remember right was the original error reported. When I did some quick checking then it didn’t seem to be a autoweka class. But it apparently got serialized into the model somehow, but doesn’t deserialize. I might look a little more.


I’m not sure what I searched last time but this time googled


It does turn up on automl. I think someone associated with that had responded earlier as well?

public class ConfigurationCollection
extends XmlSerializable

It appears to implement it’s own XML flavored serialization? 

You might check with them. Maybe raise on issue on their GitHub. If the earlier atonal poster isn’t currently following this.


One last thought that might be if they have a ‘new’ model serialization like WekaSerialConverter was written for then maybe WekaSerialConverter could be modified to convert old ones to new ones there as well. I would need more information on what they are doing to accomplish serialization. 
Again to convert the same or compatible training data needs to be available.


_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Tony D
Michael Hall wrote
>> On Mar 11, 2021, at 5:15 PM, Michael Hall &lt;

> mik3hall@

> &gt; wrote:
>>
>>>
>>> Which if I remember right was the original error reported. When I did
>>> some quick checking then it didn’t seem to be a autoweka class. But it
>>> apparently got serialized into the model somehow, but doesn’t
>>> deserialize. I might look a little more.
>>>
>>
>> I’m not sure what I searched last time but this time googled
>>
>> https://automl.github.io/autoweka/index.html?autoweka/ConfigurationCollection.html
>> &lt;https://automl.github.io/autoweka/index.html?autoweka/ConfigurationCollection.html&gt;
>>
>> It does turn up on automl. I think someone associated with that had
>> responded earlier as well?
>>
>> public class ConfigurationCollection
>> extends XmlSerializable
>>
>> It appears to implement it’s own XML flavored serialization?
>>
>> You might check with them. Maybe raise on issue on their GitHub. If the
>> earlier atonal poster isn’t currently following this.
>>
>>
> One last thought that might be if they have a ‘new’ model serialization
> like WekaSerialConverter was written for then maybe WekaSerialConverter
> could be modified to convert old ones to new ones there as well. I would
> need more information on what they are doing to accomplish serialization.
> Again to convert the same or compatible training data needs to be
> available.
>
>
> _______________________________________________
> Wekalist mailing list --

> wekalist@.ac

> Send posts to

> wekalist@.ac

> To unsubscribe send an email to

> wekalist-leave@.ac

> 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

I will have a look at the automl github for that XML version - or whatever
else might be there.

yes, you did have the original error code correct and it IS bizarre... The
model wrote to file without announced error - which only appeared when
trying to LOAD the model (which, I realized, is into a classifier that may
not be the same bloodline as the autoweka classifier, self-contained as
autoweka is...)
I wonder - if we know which weka version is embedded inside autoweka, might
it work to run that weka with the autoweka tab?...

I still haven't successfully compiled the wekaserialconverter as yet.
haven't solved the missing bits mystery. My learning to do here...




--
Sent from: https://weka.8497.n7.nabble.com/
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Michael Hall


On Mar 12, 2021, at 10:13 AM, Tony D <[hidden email]> wrote:

yes, you did have the original error code correct and it IS bizarre... The
model wrote to file without announced error 

A guess might be that AutoWeka somehow handles it’s own serialization but Weka the de-serialization.

If this project uses it’s own non-compatible serialization it won’t work

This would probably require a core Weka change. Something like

If (Object instanced Serialized)
doWekaDeserialize()
else
doAutoMLDeserialize()

Packages probably shouldn’t really require core Weka changes to work correctly.
It would be up to the Weka developers if its worth the special handling. Or figure it is worth supporting for other possible package use.


_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Tony D
Michael Hall wrote
>> On Mar 12, 2021, at 10:13 AM, Tony D &lt;

> tdecarmine@

> &gt; wrote:
>>
>> yes, you did have the original error code correct and it IS bizarre...
>> The
>> model wrote to file without announced error
>
> A guess might be that AutoWeka somehow handles it’s own serialization but
> Weka the de-serialization.
>
> If this project uses it’s own non-compatible serialization it won’t work
>
> This would probably require a core Weka change. Something like
>
> If (Object instanced Serialized)
> doWekaDeserialize()
> else
> doAutoMLDeserialize()
>
> Packages probably shouldn’t really require core Weka changes to work
> correctly.
> It would be up to the Weka developers if its worth the special handling.
> Or figure it is worth supporting for other possible package use.
>
>
>
> _______________________________________________
> Wekalist mailing list --

> wekalist@.ac

> Send posts to

> wekalist@.ac

> To unsubscribe send an email to

> wekalist-leave@.ac

> 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

I get you.

Before I get out of your hair, is it possible for you to parse the model and
pass me the argument set therein?  I may still be able to make some progress
with the model as-is until the dataset ages out. TIA...




--
Sent from: https://weka.8497.n7.nabble.com/
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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: Auto-weka problem

Michael Hall



I get you.

Before I get out of your hair, is it possible for you to parse the model and
pass me the argument set therein?  I may still be able to make some progress
with the model as-is until the dataset ages out. TIA…


My code was written to parse original Java Serialization. The model file you made available used what appears to be autumn Auto-Weka project specific XML serialization. There are I think other 3rd party XML serialization API’s available they chose not to use. 
At this point the automl people would seem to be the best ones to talk to about their project specific models.


_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
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
12