Saving/loading the trained model

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

Saving/loading the trained model

abasian
Greetings of the day

If I am using two datasets train.arff and test.arff, what is the procedure to run the classifier and get unbiased results? Do I need to follow the following steps?
(a) First I have to load the train data in the Preprocess tab
(b) Select the classifier and select " Use Training set"
(c) save the model
(d) load the test dataset in " Supplied test set"
(e) load the saved model and re-evaluate the model on test data

Please correct me if I am wrong? 

_______________________________________________
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: Saving/loading the trained model

Eibe Frank-2
Administrator

Assuming you are not modifying the data in the Preprocess panel, your steps are fine but unnecessarily complicated. You don’t need to evaluate on the training data and save the model. Just use the “Supplied test set” option immediately once you have selected a classifier to use. The classifier will be built on the data from the Preprocess panel and then evaluated on the test set.

 

If you are doing some filtering of the data, the safest way is to avoid using the Preprocess panel and to apply the FilteredClassifier instead. In particular, it is essential to avoid using a supervised filter in the Preprocess panel before evaluating a classifier in the Classify panel.  In early versions of WEKA, application of supervised filters in the Preprocess panel was actually disabled for this reason! Unsupervised filters should generally be OK to use but even with those, it is better to use FilteredClassifier so that the same preprocessing steps will be applied to the test data and the two sets of data will be consistent.

 

Note that the StringToWordVector filter is an “unsupervised” filter that is actually not strictly unsupervised because it generates a dictionary on a per-class basis before merging the two dictionaries into one. There is an option to turn this off

 

Cheers,

Eibe

 

From: [hidden email] <[hidden email]> On Behalf Of Asad Ali
Sent: Wednesday, 17 July 2019 10:29 PM
To: Weka machine learning workbench list. <[hidden email]>
Subject: [Wekalist] Saving/loading the trained model

 

Greetings of the day

 

If I am using two datasets train.arff and test.arff, what is the procedure to run the classifier and get unbiased results? Do I need to follow the following steps?

(a) First I have to load the train data in the Preprocess tab

(b) Select the classifier and select " Use Training set"

(c) save the model

(d) load the test dataset in " Supplied test set"

(e) load the saved model and re-evaluate the model on test data

 

Please correct me if I am wrong? 


_______________________________________________
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: Saving/loading the trained model

asadbtk
Hi Eide, in response to your answer in this thread, I have a question:

You wrote, "it is essential to avoid using a supervised filter in the Preprocess panel before evaluating a classifier in the Classify panel".  

What if we want to use SMOTE (supervised filter) on training data and then save the balanced data as train.arff. I mean it is not possible via FilteredClassifier

Second, you also mentioned that no need to save your model, but in the following link, Jason suggested it.


On Wed, Jul 17, 2019 at 11:14 PM <[hidden email]> wrote:

Assuming you are not modifying the data in the Preprocess panel, your steps are fine but unnecessarily complicated. You don’t need to evaluate on the training data and save the model. Just use the “Supplied test set” option immediately once you have selected a classifier to use. The classifier will be built on the data from the Preprocess panel and then evaluated on the test set.

 

If you are doing some filtering of the data, the safest way is to avoid using the Preprocess panel and to apply the FilteredClassifier instead. In particular, it is essential to avoid using a supervised filter in the Preprocess panel before evaluating a classifier in the Classify panel.  In early versions of WEKA, application of supervised filters in the Preprocess panel was actually disabled for this reason! Unsupervised filters should generally be OK to use but even with those, it is better to use FilteredClassifier so that the same preprocessing steps will be applied to the test data and the two sets of data will be consistent.

 

Note that the StringToWordVector filter is an “unsupervised” filter that is actually not strictly unsupervised because it generates a dictionary on a per-class basis before merging the two dictionaries into one. There is an option to turn this off

 

Cheers,

Eibe

 

From: [hidden email] <[hidden email]> On Behalf Of Asad Ali
Sent: Wednesday, 17 July 2019 10:29 PM
To: Weka machine learning workbench list. <[hidden email]>
Subject: [Wekalist] Saving/loading the trained model

 

Greetings of the day

 

If I am using two datasets train.arff and test.arff, what is the procedure to run the classifier and get unbiased results? Do I need to follow the following steps?

(a) First I have to load the train data in the Preprocess tab

(b) Select the classifier and select " Use Training set"

(c) save the model

(d) load the test dataset in " Supplied test set"

(e) load the saved model and re-evaluate the model on test data

 

Please correct me if I am wrong? 

_______________________________________________
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: Saving/loading the trained model

Eibe Frank-2
Administrator

You can use SMOTE in the FilteredClassifier. In that case, SMOTE will only modify the training data, not the test data. The way this works is that filters in WEKA always treat the first batch of data they receive as the training data and all subsequent batches as test data. So the first batch of data that the FilteredClassifier passes to the (SMOTE) filter, which will be the training data that comes into the FilteredClassifier, will be modified by the filter (i.e., new instances will be generated and added to the minority class) and all subsequent data will be passed through the filter untouched.

 

Another example would be a discretization filter. The discretization intervals will be found from the first batch of data that the filter receives, and all subsequent batches of data will be processed based on those intervals. So, if you perform supervised discretisation, only the training data will be used to find the discretisation intervals.

 

Regarding Jason’s article, if you want to use the model later again, you should obviously save it, but there is no need to save it if you only want to evaluate on one test set and will not need the model again.

 

Cheers,

Eibe

 

From: [hidden email] <[hidden email]> On Behalf Of javed khan
Sent: Thursday, 18 July 2019 9:35 PM
To: Weka machine learning workbench list. <[hidden email]>
Subject: Re: [Wekalist] Saving/loading the trained model

 

Hi Eide, in response to your answer in this thread, I have a question:

 

You wrote, "it is essential to avoid using a supervised filter in the Preprocess panel before evaluating a classifier in the Classify panel".  

 

What if we want to use SMOTE (supervised filter) on training data and then save the balanced data as train.arff. I mean it is not possible via FilteredClassifier

 

Second, you also mentioned that no need to save your model, but in the following link, Jason suggested it.

 

 

On Wed, Jul 17, 2019 at 11:14 PM <[hidden email]> wrote:

Assuming you are not modifying the data in the Preprocess panel, your steps are fine but unnecessarily complicated. You don’t need to evaluate on the training data and save the model. Just use the “Supplied test set” option immediately once you have selected a classifier to use. The classifier will be built on the data from the Preprocess panel and then evaluated on the test set.

 

If you are doing some filtering of the data, the safest way is to avoid using the Preprocess panel and to apply the FilteredClassifier instead. In particular, it is essential to avoid using a supervised filter in the Preprocess panel before evaluating a classifier in the Classify panel.  In early versions of WEKA, application of supervised filters in the Preprocess panel was actually disabled for this reason! Unsupervised filters should generally be OK to use but even with those, it is better to use FilteredClassifier so that the same preprocessing steps will be applied to the test data and the two sets of data will be consistent.

 

Note that the StringToWordVector filter is an “unsupervised” filter that is actually not strictly unsupervised because it generates a dictionary on a per-class basis before merging the two dictionaries into one. There is an option to turn this off

 

Cheers,

Eibe

 

From: [hidden email] <[hidden email]> On Behalf Of Asad Ali
Sent: Wednesday, 17 July 2019 10:29 PM
To: Weka machine learning workbench list. <[hidden email]>
Subject: [Wekalist] Saving/loading the trained model

 

Greetings of the day

 

If I am using two datasets train.arff and test.arff, what is the procedure to run the classifier and get unbiased results? Do I need to follow the following steps?

(a) First I have to load the train data in the Preprocess tab

(b) Select the classifier and select " Use Training set"

(c) save the model

(d) load the test dataset in " Supplied test set"

(e) load the saved model and re-evaluate the model on test data

 

Please correct me if I am wrong? 

_______________________________________________
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: Saving/loading the trained model

asadbtk
Thanks Eide again, So it means the SMOTE will be applied to the data(set) which we provide in the Pre-process section and not to the test data which we will provide in the Classify tab (i.e. Classify test set)?
Using the filter in the Preprocess section is not useful or not recommended (gives wrong/biased results)? Because I have seen people who are using pre-process section for the filtering purposes?


On Thu, Jul 18, 2019 at 12:35 PM <[hidden email]> wrote:

You can use SMOTE in the FilteredClassifier. In that case, SMOTE will only modify the training data, not the test data. The way this works is that filters in WEKA always treat the first batch of data they receive as the training data and all subsequent batches as test data. So the first batch of data that the FilteredClassifier passes to the (SMOTE) filter, which will be the training data that comes into the FilteredClassifier, will be modified by the filter (i.e., new instances will be generated and added to the minority class) and all subsequent data will be passed through the filter untouched.

 

Another example would be a discretization filter. The discretization intervals will be found from the first batch of data that the filter receives, and all subsequent batches of data will be processed based on those intervals. So, if you perform supervised discretisation, only the training data will be used to find the discretisation intervals.

 

Regarding Jason’s article, if you want to use the model later again, you should obviously save it, but there is no need to save it if you only want to evaluate on one test set and will not need the model again.

 

Cheers,

Eibe

 

From: [hidden email] <[hidden email]> On Behalf Of javed khan
Sent: Thursday, 18 July 2019 9:35 PM
To: Weka machine learning workbench list. <[hidden email]>
Subject: Re: [Wekalist] Saving/loading the trained model

 

Hi Eide, in response to your answer in this thread, I have a question:

 

You wrote, "it is essential to avoid using a supervised filter in the Preprocess panel before evaluating a classifier in the Classify panel".  

 

What if we want to use SMOTE (supervised filter) on training data and then save the balanced data as train.arff. I mean it is not possible via FilteredClassifier

 

Second, you also mentioned that no need to save your model, but in the following link, Jason suggested it.

 

 

On Wed, Jul 17, 2019 at 11:14 PM <[hidden email]> wrote:

Assuming you are not modifying the data in the Preprocess panel, your steps are fine but unnecessarily complicated. You don’t need to evaluate on the training data and save the model. Just use the “Supplied test set” option immediately once you have selected a classifier to use. The classifier will be built on the data from the Preprocess panel and then evaluated on the test set.

 

If you are doing some filtering of the data, the safest way is to avoid using the Preprocess panel and to apply the FilteredClassifier instead. In particular, it is essential to avoid using a supervised filter in the Preprocess panel before evaluating a classifier in the Classify panel.  In early versions of WEKA, application of supervised filters in the Preprocess panel was actually disabled for this reason! Unsupervised filters should generally be OK to use but even with those, it is better to use FilteredClassifier so that the same preprocessing steps will be applied to the test data and the two sets of data will be consistent.

 

Note that the StringToWordVector filter is an “unsupervised” filter that is actually not strictly unsupervised because it generates a dictionary on a per-class basis before merging the two dictionaries into one. There is an option to turn this off

 

Cheers,

Eibe

 

From: [hidden email] <[hidden email]> On Behalf Of Asad Ali
Sent: Wednesday, 17 July 2019 10:29 PM
To: Weka machine learning workbench list. <[hidden email]>
Subject: [Wekalist] Saving/loading the trained model

 

Greetings of the day

 

If I am using two datasets train.arff and test.arff, what is the procedure to run the classifier and get unbiased results? Do I need to follow the following steps?

(a) First I have to load the train data in the Preprocess tab

(b) Select the classifier and select " Use Training set"

(c) save the model

(d) load the test dataset in " Supplied test set"

(e) load the saved model and re-evaluate the model on test data

 

Please correct me if I am wrong? 

_______________________________________________
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

_______________________________________________
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: Saving/loading the trained model

Eibe Frank-2
Administrator

Yes, that’s the way the FilteredClassifier will apply SMOTE.

 

If you apply SMOTE in the Preprocess tab  and then provide a separate test set in the Classify tab, that is totally fine. The problem occurs when you use a cross-validation or a percentage split evaluation in the Classify tab based on the SMOTE-processed data from the Preprocess tab because then some of the artificial data generated by SMOTE will be used as the test data, which is problematic.

 

Using something like an unsupervised discretisation filter in the Preprocess panel is generally OK. It does not make use of the class information.

 

An extreme case of a problematic evaluation would be if you used the AddClassification filter in the Preprocess panel to add an attribute with the predicted classification and then performed a cross-validation based on that data using the predicted classification as a predictor attribute in the Classify tab.

 

A good rule of thump is to avoid using supervised filters (and StringToWordVector in default mode) in the Preprocess panel. Depending on your application, you may want to carefully think about the application of unsupervised filters there as well. An easy way around all these problems is to use the FilteredClassifier, possibly using the MultiFilter to combine multiple filters inside it.

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Friday, 19 July 2019 1:24 AM
To: [hidden email]
Subject: Re: [Wekalist] Saving/loading the trained model

 

Thanks Eide again, So it means the SMOTE will be applied to the data(set) which we provide in the Pre-process section and not to the test data which we will provide in the Classify tab (i.e. Classify test set)?

Using the filter in the Preprocess section is not useful or not recommended (gives wrong/biased results)? Because I have seen people who are using pre-process section for the filtering purposes?

 

 

On Thu, Jul 18, 2019 at 12:35 PM <[hidden email]> wrote:

You can use SMOTE in the FilteredClassifier. In that case, SMOTE will only modify the training data, not the test data. The way this works is that filters in WEKA always treat the first batch of data they receive as the training data and all subsequent batches as test data. So the first batch of data that the FilteredClassifier passes to the (SMOTE) filter, which will be the training data that comes into the FilteredClassifier, will be modified by the filter (i.e., new instances will be generated and added to the minority class) and all subsequent data will be passed through the filter untouched.

 

Another example would be a discretization filter. The discretization intervals will be found from the first batch of data that the filter receives, and all subsequent batches of data will be processed based on those intervals. So, if you perform supervised discretisation, only the training data will be used to find the discretisation intervals.

 

Regarding Jason’s article, if you want to use the model later again, you should obviously save it, but there is no need to save it if you only want to evaluate on one test set and will not need the model again.

 

Cheers,

Eibe

 

From: [hidden email] <[hidden email]> On Behalf Of javed khan
Sent: Thursday, 18 July 2019 9:35 PM
To: Weka machine learning workbench list. <[hidden email]>
Subject: Re: [Wekalist] Saving/loading the trained model

 

Hi Eide, in response to your answer in this thread, I have a question:

 

You wrote, "it is essential to avoid using a supervised filter in the Preprocess panel before evaluating a classifier in the Classify panel".  

 

What if we want to use SMOTE (supervised filter) on training data and then save the balanced data as train.arff. I mean it is not possible via FilteredClassifier

 

Second, you also mentioned that no need to save your model, but in the following link, Jason suggested it.

 

 

On Wed, Jul 17, 2019 at 11:14 PM <[hidden email]> wrote:

Assuming you are not modifying the data in the Preprocess panel, your steps are fine but unnecessarily complicated. You don’t need to evaluate on the training data and save the model. Just use the “Supplied test set” option immediately once you have selected a classifier to use. The classifier will be built on the data from the Preprocess panel and then evaluated on the test set.

 

If you are doing some filtering of the data, the safest way is to avoid using the Preprocess panel and to apply the FilteredClassifier instead. In particular, it is essential to avoid using a supervised filter in the Preprocess panel before evaluating a classifier in the Classify panel.  In early versions of WEKA, application of supervised filters in the Preprocess panel was actually disabled for this reason! Unsupervised filters should generally be OK to use but even with those, it is better to use FilteredClassifier so that the same preprocessing steps will be applied to the test data and the two sets of data will be consistent.

 

Note that the StringToWordVector filter is an “unsupervised” filter that is actually not strictly unsupervised because it generates a dictionary on a per-class basis before merging the two dictionaries into one. There is an option to turn this off

 

Cheers,

Eibe

 

From: [hidden email] <[hidden email]> On Behalf Of Asad Ali
Sent: Wednesday, 17 July 2019 10:29 PM
To: Weka machine learning workbench list. <[hidden email]>
Subject: [Wekalist] Saving/loading the trained model

 

Greetings of the day

 

If I am using two datasets train.arff and test.arff, what is the procedure to run the classifier and get unbiased results? Do I need to follow the following steps?

(a) First I have to load the train data in the Preprocess tab

(b) Select the classifier and select " Use Training set"

(c) save the model

(d) load the test dataset in " Supplied test set"

(e) load the saved model and re-evaluate the model on test data

 

Please correct me if I am wrong? 

_______________________________________________
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

 


_______________________________________________
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: Saving/loading the trained model

asadbtk
Thanks once again Eibe, for your detailed explanation. 



On Fri, Jul 19, 2019 at 12:51 AM Eibe Frank <[hidden email]> wrote:

Yes, that’s the way the FilteredClassifier will apply SMOTE.

 

If you apply SMOTE in the Preprocess tab  and then provide a separate test set in the Classify tab, that is totally fine. The problem occurs when you use a cross-validation or a percentage split evaluation in the Classify tab based on the SMOTE-processed data from the Preprocess tab because then some of the artificial data generated by SMOTE will be used as the test data, which is problematic.

 

Using something like an unsupervised discretisation filter in the Preprocess panel is generally OK. It does not make use of the class information.

 

An extreme case of a problematic evaluation would be if you used the AddClassification filter in the Preprocess panel to add an attribute with the predicted classification and then performed a cross-validation based on that data using the predicted classification as a predictor attribute in the Classify tab.

 

A good rule of thump is to avoid using supervised filters (and StringToWordVector in default mode) in the Preprocess panel. Depending on your application, you may want to carefully think about the application of unsupervised filters there as well. An easy way around all these problems is to use the FilteredClassifier, possibly using the MultiFilter to combine multiple filters inside it.

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Friday, 19 July 2019 1:24 AM
To: [hidden email]
Subject: Re: [Wekalist] Saving/loading the trained model

 

Thanks Eide again, So it means the SMOTE will be applied to the data(set) which we provide in the Pre-process section and not to the test data which we will provide in the Classify tab (i.e. Classify test set)?

Using the filter in the Preprocess section is not useful or not recommended (gives wrong/biased results)? Because I have seen people who are using pre-process section for the filtering purposes?

 

 

On Thu, Jul 18, 2019 at 12:35 PM <[hidden email]> wrote:

You can use SMOTE in the FilteredClassifier. In that case, SMOTE will only modify the training data, not the test data. The way this works is that filters in WEKA always treat the first batch of data they receive as the training data and all subsequent batches as test data. So the first batch of data that the FilteredClassifier passes to the (SMOTE) filter, which will be the training data that comes into the FilteredClassifier, will be modified by the filter (i.e., new instances will be generated and added to the minority class) and all subsequent data will be passed through the filter untouched.

 

Another example would be a discretization filter. The discretization intervals will be found from the first batch of data that the filter receives, and all subsequent batches of data will be processed based on those intervals. So, if you perform supervised discretisation, only the training data will be used to find the discretisation intervals.

 

Regarding Jason’s article, if you want to use the model later again, you should obviously save it, but there is no need to save it if you only want to evaluate on one test set and will not need the model again.

 

Cheers,

Eibe

 

From: [hidden email] <[hidden email]> On Behalf Of javed khan
Sent: Thursday, 18 July 2019 9:35 PM
To: Weka machine learning workbench list. <[hidden email]>
Subject: Re: [Wekalist] Saving/loading the trained model

 

Hi Eide, in response to your answer in this thread, I have a question:

 

You wrote, "it is essential to avoid using a supervised filter in the Preprocess panel before evaluating a classifier in the Classify panel".  

 

What if we want to use SMOTE (supervised filter) on training data and then save the balanced data as train.arff. I mean it is not possible via FilteredClassifier

 

Second, you also mentioned that no need to save your model, but in the following link, Jason suggested it.

 

 

On Wed, Jul 17, 2019 at 11:14 PM <[hidden email]> wrote:

Assuming you are not modifying the data in the Preprocess panel, your steps are fine but unnecessarily complicated. You don’t need to evaluate on the training data and save the model. Just use the “Supplied test set” option immediately once you have selected a classifier to use. The classifier will be built on the data from the Preprocess panel and then evaluated on the test set.

 

If you are doing some filtering of the data, the safest way is to avoid using the Preprocess panel and to apply the FilteredClassifier instead. In particular, it is essential to avoid using a supervised filter in the Preprocess panel before evaluating a classifier in the Classify panel.  In early versions of WEKA, application of supervised filters in the Preprocess panel was actually disabled for this reason! Unsupervised filters should generally be OK to use but even with those, it is better to use FilteredClassifier so that the same preprocessing steps will be applied to the test data and the two sets of data will be consistent.

 

Note that the StringToWordVector filter is an “unsupervised” filter that is actually not strictly unsupervised because it generates a dictionary on a per-class basis before merging the two dictionaries into one. There is an option to turn this off

 

Cheers,

Eibe

 

From: [hidden email] <[hidden email]> On Behalf Of Asad Ali
Sent: Wednesday, 17 July 2019 10:29 PM
To: Weka machine learning workbench list. <[hidden email]>
Subject: [Wekalist] Saving/loading the trained model

 

Greetings of the day

 

If I am using two datasets train.arff and test.arff, what is the procedure to run the classifier and get unbiased results? Do I need to follow the following steps?

(a) First I have to load the train data in the Preprocess tab

(b) Select the classifier and select " Use Training set"

(c) save the model

(d) load the test dataset in " Supplied test set"

(e) load the saved model and re-evaluate the model on test data

 

Please correct me if I am wrong? 

_______________________________________________
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

 

_______________________________________________
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