classifier for non-linear relationships

[Michael Michael <mi...@cs.stanford.edu>]

Hi everyone,

 

I am new to the forum and mahout (I apologize for the newbieness of my
questions if they have been answered in the past) and I was wondering if
there are any implementations out there of a classifier that can handle
highly non-linear relationships and provide a binary output. Currently, we
are using the kernel-adatron algorithm in the neurosolutions commercial
package and it works very well for us. Neurosolutions was fast to train and
provided better results than anything else that we tested and had
exceptionally high training and recall speed.  However, the license is very
expensive if you want to scale to 10's of computers operating simultaneously
for a few minutes a day for recall. 

 

Are there any other implementations out there that can help us out? I
appreciate any pointers of help in advance.

 

Cheers,

 

m2TM

 

Re: classifier for non-linear relationships

[Ted Dunning <te...@gmail.com>]

Some of each.

There is now some momentum for moving the random forest implementation
forward a bit.

But the real problem is that to really develop a solution for this takes
quite a bit of effort.  Supporting it takes more.

If you have a partial or full solution to the resource issue, that would be
of great interest.

On Thu, Mar 28, 2013 at 6:35 PM, Ray <rt...@gmail.com> wrote:

> Why does Mahout not have what this person wants?
>
> Is it not really in the scope of Mahout?  Used to be in Mahout, gone now?
>  Needed someone to put it there in the first place?
>
> Ray.
>
>
>
> On 03/28/2013 07:02 AM, Michael Michael wrote:
>
>> Thanks Ted. I contacted them to find out about pricing, but I am sure it
>> will be expensive though.
>>
>> It seems that since there are no open source solutions on this, my best
>> is either matlab or to purchase something from a company like
>> neurosolutions or skytree (there are a few others that fit the bill as
>> well).
>>
>> Cheers,
>>
>> m2TM
>>
>> -----Original Message-----
>> From: Ted Dunning [mailto:ted.dunning@gmail.com]
>> Sent: Wednesday, March 27, 2013 12:11 PM
>> To: user@mahout.apache.org
>> Subject: Re: classifier for non-linear relationships
>>
>> SkyTree comes to mind.
>>
>> Mahout doesn't have what you want.
>>
>> On Wed, Mar 27, 2013 at 3:28 PM, Michael Michael
>> <mi...@cs.stanford.edu>**wrote:
>>
>>  Hi everyone,
>>>
>>>
>>>
>>> I am new to the forum and mahout (I apologize for the newbieness of my
>>> questions if they have been answered in the past) and I was wondering
>>> if there are any implementations out there of a classifier that can
>>> handle highly non-linear relationships and provide a binary output.
>>> Currently, we are using the kernel-adatron algorithm in the
>>> neurosolutions commercial package and it works very well for us.
>>> Neurosolutions was fast to train and provided better results than
>>> anything else that we tested and had exceptionally high training and
>>> recall speed.  However, the license is very expensive if you want to
>>> scale to 10's of computers operating simultaneously for a few minutes
>>> a day for recall.
>>>
>>>
>>>
>>> Are there any other implementations out there that can help us out? I
>>> appreciate any pointers of help in advance.
>>>
>>>
>>>
>>> Cheers,
>>>
>>>
>>>
>>> m2TM
>>>
>>>
>>>
>>>
>>>
>

Re: classifier for non-linear relationships

[Ray <rt...@gmail.com>]

Why does Mahout not have what this person wants?

Is it not really in the scope of Mahout?  Used to be in Mahout, gone 
now?  Needed someone to put it there in the first place?

Ray.


On 03/28/2013 07:02 AM, Michael Michael wrote:
> Thanks Ted. I contacted them to find out about pricing, but I am sure it will be expensive though.
>
> It seems that since there are no open source solutions on this, my best is either matlab or to purchase something from a company like neurosolutions or skytree (there are a few others that fit the bill as well).
>
> Cheers,
>
> m2TM
>
> -----Original Message-----
> From: Ted Dunning [mailto:ted.dunning@gmail.com]
> Sent: Wednesday, March 27, 2013 12:11 PM
> To: user@mahout.apache.org
> Subject: Re: classifier for non-linear relationships
>
> SkyTree comes to mind.
>
> Mahout doesn't have what you want.
>
> On Wed, Mar 27, 2013 at 3:28 PM, Michael Michael
> <mi...@cs.stanford.edu>wrote:
>
>> Hi everyone,
>>
>>
>>
>> I am new to the forum and mahout (I apologize for the newbieness of my
>> questions if they have been answered in the past) and I was wondering
>> if there are any implementations out there of a classifier that can
>> handle highly non-linear relationships and provide a binary output.
>> Currently, we are using the kernel-adatron algorithm in the
>> neurosolutions commercial package and it works very well for us.
>> Neurosolutions was fast to train and provided better results than
>> anything else that we tested and had exceptionally high training and
>> recall speed.  However, the license is very expensive if you want to
>> scale to 10's of computers operating simultaneously for a few minutes
>> a day for recall.
>>
>>
>>
>> Are there any other implementations out there that can help us out? I
>> appreciate any pointers of help in advance.
>>
>>
>>
>> Cheers,
>>
>>
>>
>> m2TM
>>
>>
>>
>>

RE: classifier for non-linear relationships

[Michael Michael <mi...@cs.stanford.edu>]

Thanks Ted. I contacted them to find out about pricing, but I am sure it will be expensive though.

It seems that since there are no open source solutions on this, my best is either matlab or to purchase something from a company like neurosolutions or skytree (there are a few others that fit the bill as well).

Cheers,

m2TM

-----Original Message-----
From: Ted Dunning [mailto:ted.dunning@gmail.com] 
Sent: Wednesday, March 27, 2013 12:11 PM
To: user@mahout.apache.org
Subject: Re: classifier for non-linear relationships

SkyTree comes to mind.

Mahout doesn't have what you want.

On Wed, Mar 27, 2013 at 3:28 PM, Michael Michael
<mi...@cs.stanford.edu>wrote:

> Hi everyone,
>
>
>
> I am new to the forum and mahout (I apologize for the newbieness of my 
> questions if they have been answered in the past) and I was wondering 
> if there are any implementations out there of a classifier that can 
> handle highly non-linear relationships and provide a binary output. 
> Currently, we are using the kernel-adatron algorithm in the 
> neurosolutions commercial package and it works very well for us. 
> Neurosolutions was fast to train and provided better results than 
> anything else that we tested and had exceptionally high training and 
> recall speed.  However, the license is very expensive if you want to 
> scale to 10's of computers operating simultaneously for a few minutes 
> a day for recall.
>
>
>
> Are there any other implementations out there that can help us out? I 
> appreciate any pointers of help in advance.
>
>
>
> Cheers,
>
>
>
> m2TM
>
>
>
>

Re: classifier for non-linear relationships

[Ted Dunning <te...@gmail.com>]

SkyTree comes to mind.

Mahout doesn't have what you want.

On Wed, Mar 27, 2013 at 3:28 PM, Michael Michael
<mi...@cs.stanford.edu>wrote:

> Hi everyone,
>
>
>
> I am new to the forum and mahout (I apologize for the newbieness of my
> questions if they have been answered in the past) and I was wondering if
> there are any implementations out there of a classifier that can handle
> highly non-linear relationships and provide a binary output. Currently, we
> are using the kernel-adatron algorithm in the neurosolutions commercial
> package and it works very well for us. Neurosolutions was fast to train and
> provided better results than anything else that we tested and had
> exceptionally high training and recall speed.  However, the license is very
> expensive if you want to scale to 10's of computers operating
> simultaneously
> for a few minutes a day for recall.
>
>
>
> Are there any other implementations out there that can help us out? I
> appreciate any pointers of help in advance.
>
>
>
> Cheers,
>
>
>
> m2TM
>
>
>
>