dials 1.4.4

• attention_type(), dropout_hidden(), dropout_last(), and target_token() are new parameters for the SAINT model in the brulee package (#452).

• bottleneck_units(), dropout_attn(), dropout_embedding(), num_attn_blocks(), num_attn_feat(), num_attn_heads(), num_embedding(), penalty_average(), penalty_type(), normalization(), resid_at(), and step_rate() are new parameters for deep learning models in the tabular package when fit with the brulee engine (#452, #461).

• l2_leaf_reg() and max_leaves() are new parameter objects for tree-based models that use the "catboost" engine in the bonsai package (#453).

• splitting_rule() gains support for survival forests (#456).