Re: [TMVA-users] BDT training with negative weights.

[Helge V. <Hel...@ce...>]

Yes, bagging or not bagging doesn't matter. Just AdaBoost (with or
without additional bagging) doesn't handle negative weights, but
GradBoost seems to do.

Cheers,

Helge


On 18 January 2016 at 18:04, Per Olov Joakim Gradin
<per...@ce...> wrote:
> Hi Helge,
>
> Great, I will try that option. Does the same apply if we are using bagged boost?
>
> Thanks,
> Joakim
> ________________________________________
> From: hel...@gm... [hel...@gm...] on behalf of Helge Voss [Hel...@ce...]
> Sent: 18 January 2016 17:54
> To: Per Olov Joakim Gradin
> Cc: tmv...@li...; Max Fredrik Isacson; Elin Bergeaas Kuutmann; Hadar Cohen
> Subject: Re: [TMVA-users] BDT training with negative weights.
>
> Actually, for 'Gradient Boosting' I was recently shown that the 'Pray'
> option (i.e. using the standard Gradient Boost technique simply on the
> negetive event weights)
> should be working. I always planned to make more tests on that and
> 'make a public announcement' etc.. but never got around to do so .
>
> So pleaee, try 'Pray' :)
>
> Helge
>
>
> On 18 January 2016 at 16:27, Per Olov Joakim Gradin
> <per...@ce...> wrote:
>> Hi TMVA-users,
>>
>> Given Monte Carlo data with positive & negative weights, what is your
>> suggested option for training a BDT? We have previously been using gradient
>> boosted trees due to them being advertised as more robust compared to
>> adaptive boosting. However, the "InverseBoostNegWeights" option is not
>> usable with gradient boosting (in TMVA version 4.2.0), so we fall back to
>> ignoring the events with negative weight in the training.
>>
>> --- <WARNING> MyBDT                 : the option *InverseBoostNegWeights*
>> does not exist for BoostType=Grad --> change to *IgnoreNegWeightsInTraining*
>>
>> Our MC production team is strongly advising us to not drop these events and
>> that means that we have to change our training method/parameters. I am
>> wondering if either of the other two possibilities, PairNegWeightsGlobal and
>> Pray, would be recommended? If I understand it right, the Pray option is
>> using the event weights as is, are there any dangers or negative effects of
>> doing that? (I'm guessing that there are due to its name...)
>>
>> If you don't think it is wise to use one of those options, which boosting
>> type would you recommend we use instead of gradient boost?
>>
>> Cheers,
>> Joakim
>>
>> ------------------------------------------------------------------------------
>> Site24x7 APM Insight: Get Deep Visibility into Application Performance
>> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
>> Monitor end-to-end web transactions and take corrective actions now
>> Troubleshoot faster and improve end-user experience. Signup Now!
>> http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140
>> _______________________________________________
>> TMVA-users mailing list
>> TMV...@li...
>> https://lists.sourceforge.net/lists/listinfo/tmva-users
>>
>