[prev in list] [next in list] [prev in thread] [next in thread] 

List:       wekalist
Subject:    Re: [Wekalist] DI4jclassifier parameters
From:       Eibe Frank <eibe.frank () waikato ! ac ! nz>
Date:       2018-07-17 9:31:39
Message-ID: 737e235da5dbfc48f2743e0d345228f6 () mail ! gmail ! com
[Download RAW message or body]

Make sure you increase the batch size in the instance iterator. The
default is 1, which does not seem to work well with the default optimiser.
It also won't be very efficient (mini-batches of size 1 are generally not
ideal).

You should also change the loss function in the output layer to something
suitable for regression, e.g., MSE.

Finally, your NaNs indicate numerical problems. Try standardizing the
entire data, including the target attribute, in the Preprocess panel
before you switch to the Classify panel. (You can standardise the
target/class attribute by selecting "No class" above the histogram in the
Preprocess panel.)

Cheers,
Eibe

> -----Original Message-----
> From: wekalist-bounces@list.waikato.ac.nz <wekalist-
> bounces@list.waikato.ac.nz> On Behalf Of Bill Bane
> Sent: Tuesday, 17 July 2018 3:19 AM
> To: wekalist@list.waikato.ac.nz
> Subject: Re: [Wekalist] DI4jclassifier parameters
>
> Thanks, this is helpful.
>
> But I think my problem is adapting it to work with regression problems
> (numeric classes).  To illustrate, using your guidance I was able to
solve the Iris
> classification, e.g.:
>
> === Confusion Matrix ===
>
>   a  b  c   <-- classified as
>  49  1  0 |  a = Iris-setosa
>   0 45  5 |  b = Iris-versicolor
>   0  4 46 |  c = Iris-virginica
>
> However, using any numeric-class data, e.g. CPU, I get results like
this:
>
> === Cross-validation ===
> === Summary ===
>
> Correlation coefficient               NaN
> Mean absolute error                  NaN
> Root mean squared error           NaN
> Relative absolute error               NaN      %
> Root relative squared error         NaN      %
> UnClassified Instances                209              100      %
> Total Number of Instances           209
>
> The examples in the user guide (https://deeplearning.cms.waikato.ac.nz/)
are
> focused on classification problems.
>
>
>
> --
> Sent from: http://weka.8497.n7.nabble.com/
> _______________________________________________
> Wekalist mailing list
> Send posts to: Wekalist@list.waikato.ac.nz List info and subscription
status:
> 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: Wekalist@list.waikato.ac.nz
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
[prev in list] [next in list] [prev in thread] [next in thread]