You are viewing a plain text version of this content. The canonical link for it is here.

Posted to user@spark.apache.org by Hayri Volkan Agun <vo...@gmail.com> on 2016/02/08 20:02:02 UTC

Spark LBFGS Error with ANN

I am using Multilayer Percertron Classifier. In each training instance
there are multiple 1.0 in the ouput vector of the Multilayer Perceptron
Classifier. This is necessary. With small number of training data I am
getting the following error

*ERROR LBFGS: Failure again! Giving up and returning. Maybe the objective
is just poorly behaved? *


What is the problem? My objective output is a vector for instance :

[1.0 0.0 0.0 1.0] is my label with 4 double values. My network topology is
100,20,4.

Thanks in advance
Hayri Volkan Agun
PhD. Student - Anadolu University

Re: Spark LBFGS Error with ANN

Posted by Hayri Volkan Agun <vo...@gmail.com>.

Hi Alexander,

Thanks for the info.
I modified the code and used sigmoid at latest layer. It worked correctly
with 2-3 layer.

Thanks...


On Tue, Feb 16, 2016 at 8:51 PM, Ulanov, Alexander <alexander.ulanov@hpe.com
> wrote:

> Hi Hayri,
>
>
>
> The MLP classifier is multi-class (one class per instance) but not
> multi-label (multiple classes per instance). The top layer of the network
> is softmax
> http://spark.apache.org/docs/latest/ml-classification-regression.html#multilayer-perceptron-classifier
> that requires the outputs sum to 1. Public interface accepts vector-label
> pairs.
>
>
>
> Best regards, Alexander
>
>
>
> *From:* Hayri Volkan Agun [mailto:volkanagun@gmail.com]
> *Sent:* Monday, February 08, 2016 11:02 AM
> *To:* user @spark
> *Subject:* Spark LBFGS Error with ANN
>
>
>
> I am using Multilayer Percertron Classifier. In each training instance
> there are multiple 1.0 in the ouput vector of the Multilayer Perceptron
> Classifier. This is necessary. With small number of training data I am
> getting the following error
>
>
>
> *ERROR LBFGS: Failure again! Giving up and returning. Maybe the objective
> is just poorly behaved? *
>
>
>
>
>
> What is the problem? My objective output is a vector for instance :
>
>
>
> [1.0 0.0 0.0 1.0] is my label with 4 double values. My network topology is
> 100,20,4.
>
>
>
> Thanks in advance
>
> Hayri Volkan Agun
>
> PhD. Student - Anadolu University
>



-- 
Hayri Volkan Agun
PhD. Student - Anadolu University

RE: Spark LBFGS Error with ANN

Posted by "Ulanov, Alexander" <al...@hpe.com>.

Hi Hayri,

The MLP classifier is multi-class (one class per instance) but not multi-label (multiple classes per instance). The top layer of the network is softmax http://spark.apache.org/docs/latest/ml-classification-regression.html#multilayer-perceptron-classifier that requires the outputs sum to 1. Public interface accepts vector-label pairs.

Best regards, Alexander

From: Hayri Volkan Agun [mailto:volkanagun@gmail.com]
Sent: Monday, February 08, 2016 11:02 AM
To: user @spark
Subject: Spark LBFGS Error with ANN

I am using Multilayer Percertron Classifier. In each training instance there are multiple 1.0 in the ouput vector of the Multilayer Perceptron Classifier. This is necessary. With small number of training data I am getting the following error

ERROR LBFGS: Failure again! Giving up and returning. Maybe the objective is just poorly behaved?

What is the problem? My objective output is a vector for instance :

[1.0 0.0 0.0 1.0] is my label with 4 double values. My network topology is 100,20,4.

Thanks in advance
Hayri Volkan Agun
PhD. Student - Anadolu University