From 59a395ab1b596be0782b22a6901de86ef18d0989 Mon Sep 17 00:00:00 2001
From: Richard Stotz <richardstotz@google.com>
Date: Mon, 3 Jul 2023 08:46:22 -0700
Subject: [PATCH] Release TF-DF 1.4.0 and YDF 1.5.0

PiperOrigin-RevId: 545230857
---
 CHANGELOG.md                          | 11 +++++++++++
 documentation/rtd/hyper_parameters.md |  6 +++---
 2 files changed, 14 insertions(+), 3 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 167b03a6..9f15fc8d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,16 +2,27 @@
 
 ## HEAD
 
+## 1.5.0 - 2023-07-03
+
 ### Feature
 
 -   Rename experimental_analyze_model_and_dataset to analyze_model_and_dataset
 -   Add new GBT loss function `POISSON` for Poisson log likelihood.
 -   Go API: Categorical string values available for inspection.
 -   Improved training speed for unit-weight datasets.
+-   Support for MHLD oblique decision trees.
+-   Multi-threaded RMSE computation.
+-   Added Uint8 inference engine.
+-   Added Multi-task learning where the output of models trained as "secondary"
+    are used as input for the models trained as "primary"
 
 ### Fix
 
 -   Go API: fixed typo on OutOfVocabulary constant.
+-   Error messages for Uplift models.
+-   Remove owner leakage in the model compiler.
+-   Fix buggy restriction for SelGB sampling
+-   Improve documentation.
 
 ## 1.4.0 - 2023-03-20
 
diff --git a/documentation/rtd/hyper_parameters.md b/documentation/rtd/hyper_parameters.md
index 3d442089..d774c920 100644
--- a/documentation/rtd/hyper_parameters.md
+++ b/documentation/rtd/hyper_parameters.md
@@ -283,9 +283,9 @@ reasonable time.
 
 -   **Type:** Categorical **Default:** DEFAULT **Possible values:** DEFAULT,
     BINOMIAL_LOG_LIKELIHOOD, SQUARED_ERROR, MULTINOMIAL_LOG_LIKELIHOOD,
-    LAMBDA_MART_NDCG5, XE_NDCG_MART, BINARY_FOCAL_LOSS
+    LAMBDA_MART_NDCG5, XE_NDCG_MART, BINARY_FOCAL_LOSS, POISSON
 
--   The loss optimized by the model. If not specified (DEFAULT) the loss is selected automatically according to the \"task\" and label statistics. For example, if task=CLASSIFICATION and the label has two possible values, the loss will be set to BINOMIAL_LOG_LIKELIHOOD. Possible values are:<br>- `DEFAULT`: Select the loss automatically according to the task and label statistics.<br>- `BINOMIAL_LOG_LIKELIHOOD`: Binomial log likelihood. Only valid for binary classification.<br>- `SQUARED_ERROR`: Least square loss. Only valid for regression.<br>- `MULTINOMIAL_LOG_LIKELIHOOD`: Multinomial log likelihood i.e. cross-entropy. Only valid for binary or multi-class classification.<br>- `LAMBDA_MART_NDCG5`: LambdaMART with NDCG5.<br>- `XE_NDCG_MART`:  Cross Entropy Loss NDCG. See arxiv.org/abs/1911.09798.<br>
+-   The loss optimized by the model. If not specified (DEFAULT) the loss is selected automatically according to the \"task\" and label statistics. For example, if task=CLASSIFICATION and the label has two possible values, the loss will be set to BINOMIAL_LOG_LIKELIHOOD. Possible values are:<br>- `DEFAULT`: Select the loss automatically according to the task and label statistics.<br>- `BINOMIAL_LOG_LIKELIHOOD`: Binomial log likelihood. Only valid for binary classification.<br>- `SQUARED_ERROR`: Least square loss. Only valid for regression.<br>- `POISSON`: Poisson log likelihood loss. Mainly used for counting problems. Only valid for regression.<br>- `MULTINOMIAL_LOG_LIKELIHOOD`: Multinomial log likelihood i.e. cross-entropy. Only valid for binary or multi-class classification.<br>- `LAMBDA_MART_NDCG5`: LambdaMART with NDCG5.<br>- `XE_NDCG_MART`:  Cross Entropy Loss NDCG. See arxiv.org/abs/1911.09798.<br>
 
 #### [max_depth](https://github.com/google/yggdrasil-decision-forests/blob/main/yggdrasil_decision_forests/learner/decision_tree/decision_tree.proto)
 
@@ -379,7 +379,7 @@ reasonable time.
 #### [sampling_method](https://github.com/google/yggdrasil-decision-forests/blob/main/yggdrasil_decision_forests/learner/gradient_boosted_trees/gradient_boosted_trees.proto)
 
 -   **Type:** Categorical **Default:** RANDOM **Possible values:** NONE, RANDOM,
-    GOSS
+    GOSS, SELGB
 
 -   Control the sampling of the datasets used to train individual trees.<br>- NONE: No sampling is applied. This is equivalent to RANDOM sampling with \"subsample=1\".<br>- RANDOM (default): Uniform random sampling. Automatically selected if "subsample" is set.<br>- GOSS: Gradient-based One-Side Sampling. Automatically selected if "goss_alpha" or "goss_beta" is set.<br>- SELGB: Selective Gradient Boosting. Automatically selected if "selective_gradient_boosting_ratio" is set. Only valid for ranking.<br>