v1.6.4

• add likelihood = ‘zero_inflated_gamma’
• add likelihood = ‘zero_censored_power_transformed_normal’ for modeling data with a point mass at 0 and a continuous distribution for y > 0
• [R-package] partial dependence plots are now done by default on the latent predictor scale, not on the response variable scale. Can be controlled by the option 'latent_scale' (=default) of the 'gpb.plot.partial.dependence' function
• expose ‘m_lbfgs’ (number of corrections to approximate the inverse Hessian matrix for the lbfgs optimizer) and ‘delta_conv_mode_finding’ (convergence tolerance in mode finding algorithm for Laplace approximation for non-Gaussian likelihoods) to API. These can be set via the ‘params’ argument in the ‘fit’ and ‘set_optim_params’ functions
• reuse previous regression coefficients when continuing training with linear models
• offset is saved to file now
• move calculation of standard errors to ‘summary()’ and ‘get_cov_pars()’ & ‘get_coef()’ functions, discontinue ‘std_dev’ parameter