Ok thanks for the answer. There's another thing that I would like to
undestand better.
In the case I would like to change my model without retraining it, for
instance a SGD classifer, the only knobs I could turn are the ones
regarding the learning rate of the model (alpha, decayExponent etc..)
or there are other parametres?

On 15 December 2011 15:08, Ted Dunning <[email protected]> wrote:
> With the online classifiers, it is quite possible.  With the NB it is
> possible in theory, I am unsure about whether it is possible in practice.
>
> It is generally not a great idea to do this for the simple reason that you
> typically want to emphasize recent history and eliminate ancient history.
>  Controlling this to the degree you probably would like is definitely
> easier if you simply retrain your model from scratch.
>
> Moreover, the cost of retraining is typically not that high.
>
> On Thu, Dec 15, 2011 at 9:02 AM, Daniele Volpi <[email protected]>wrote:
>
>> I don't have a particular classifier model in my mind.
>> I'm just wondering if it could be possible to do.
>>
>> On 15 December 2011 13:16, Ted Dunning <[email protected]> wrote:
>> > Which kind of model?
>> >
>> > On Thu, Dec 15, 2011 at 4:41 AM, Daniele Volpi <[email protected]
>> >wrote:
>> >
>> >> During my research on classification with Mahout I've run into this
>> >> problem:
>> >> is there a way to make changes to a classifier model without the need
>> >> for retraining it?
>> >> I really appreciate any help in this regard.
>> >> Thanks
>> >> Daniele
>> >>
>>

Reply via email to