Accuracy of distribution ='Gamma', 'Tweedie', 'Gaussian'

[FelixR21]

Hi,

Just been trying this package as gbm lacks support for tweedie distribution. Have been suffering a lack of predictive power for the distributions gamma, tweedie and gaussian. I can't share the data but i've gone from a gini of 0.2 on gbm using a lognormal distribution to negative for the three listed above using gbm3.

Its the same dataset and the code has been modified to have the same parameters, any ideas what could be driving this.

thanks

Felix