Which loss functions does h2o.gbm use by default? The GBM implementation of the h2o package only allows the user to specify a loss function via the distribution argument, which defaults to multinomial for categorical response variables and gaussian for numerical response variables. According to the documentation, the loss functions are implied by the distributions. But I need to know which loss functions are used, and I can't find that anywhere in the documentation. I'm guessing it's the MSE for gaussian and cross-entropy for multinomial - does anybody here know if I'm right?
 A: GBM most likely selects the distribution based on the problem type (regression or classification) and based on the target feature:
Copied from their website:
For Classification problems:

    Bernoulli and Quasibinomial distributions are used for binary outcomes.

    A Multinomial distribution can handle multiple discrete outcomes.

For Regression problems:

    A Gaussian distribution is the function for continuous targets.

    A Poisson distribution is used for estimating counts.

    A Gamma distribution is used for estimating total values (such as claim payouts, rainfall, etc.).

    A Tweedie distribution is used for estimating densities.

    A Laplacian loss function (absolute L1-loss function) can predict the median percentile.

    A Quantile regression loss function can predict a specified percentile.

    A Huber loss function, a combination of squared error and absolute error, is more robust to outliers than L2 squared-loss function.

A: Yes, you're correct, MSE or log-loss are used by default.
There's a partial list in the documentation (read the first couple paragraphs on this distributions page too):
https://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/algo-params/distribution.html#equations
Source code:
https://github.com/h2oai/h2o-3/blob/99e5989e00c1298e3f074a8080aed5b22e2619c1/h2o-core/src/main/java/hex/DistributionFactory.java#L115
