From de8107e350f2d8ef96baf675e005b1e6aa5cf3d2 Mon Sep 17 00:00:00 2001 From: Lisa Cawley Date: Fri, 15 Nov 2019 09:36:39 -0800 Subject: [PATCH] [DOCS] Adds ml-cpp PRs to release notes (#49185) --- docs/reference/release-notes/7.5.asciidoc | 24 ++++++++++++++++++++--- 1 file changed, 21 insertions(+), 3 deletions(-) diff --git a/docs/reference/release-notes/7.5.asciidoc b/docs/reference/release-notes/7.5.asciidoc index 4caf6a9381e..52a7e037833 100644 --- a/docs/reference/release-notes/7.5.asciidoc +++ b/docs/reference/release-notes/7.5.asciidoc @@ -154,7 +154,24 @@ Machine Learning:: * Add lazy assignment job config option {pull}47726[#47726] * Additional outlier detection parameters {pull}47600[#47600] * More accurate job memory overhead {pull}47516[#47516] -* Throttle the delete-by-query of expired results {pull}47177[#47177] (issues: {issue}47003[#47003], {issue}47103[#47103]) +* Throttle the delete-by-query of expired results {pull}47177[#47177] (issue: {issue}47003[#47003]) +* Improve performance and concurrency training boosted tree regression models. +For large data sets, this change was observed to give a 10% to 20% decrease in +train time. {ml-pull}622[#622] +* Upgrade Boost libraries to version 1.71 {ml-pull}638[#638] +* Improve initialisation of boosted tree training. This generally enables us to +find lower loss models faster. {ml-pull}686[#686] +* Include a smooth tree depth based penalty to regularized objective function for +boosted tree training. Hard depth based regularization is often the strategy of +choice to prevent over fitting for XGBoost. By smoothing, we can make better tradeoffs. +Also, the parameters of the penalty function are more suited to optimising with our +Bayesian optimisation based hyperparameter search. {ml-pull}698[#698] +* Binomial logistic regression targeting cross entropy {ml-pull}713[#713] +* Improvements to count and sum anomaly detection for sparse data. This primarily +aims to improve handling of data which are predictably present: detecting when they +are unexpectedly missing. {ml-pull}721[#721] +* Trap numeric errors causing bad hyperparameter search initialisation and repeated +errors to be logged during boosted tree training {ml-pull}732[#732] Mapping:: * Add migration tool checks for _field_names disabling {pull}46972[#46972] (issues: {issue}42854[#42854], {issue}46681[#46681]) @@ -317,12 +334,14 @@ MULTIPLE AREA LABELS:: * Fix cluster alert for watcher/monitoring IndexOutOfBoundsExcep… {pull}45308[#45308] (issue: {issue}43184[#43184]) Machine Learning:: -* Deduplicate multi-fields for data frame analytics {pull}48799[#48799] (issues: {issue}48756[#48756], {issue}48770[#48770]) +* Deduplicate multi-fields for data frame analytics {pull}48799[#48799] (issue: {issue}48756[#48756]) * Prevent fetching multi-field from source {pull}48770[#48770] (issue: {issue}48756[#48756]) * Fix detection of syslog-like timestamp in find_file_structure {pull}47970[#47970] * Fix serialization of evaluation response. {pull}47557[#47557] * Reinstate ML daily maintenance actions {pull}47103[#47103] (issue: {issue}47003[#47003]) * Fix two datafeed flush lockup bugs {pull}46982[#46982] +* Restore from checkpoint could damage seasonality modeling. For example, it could +cause seasonal components to be overwritten in error. {ml-pull}821[#821] NOT CLASSIFIED:: * Remove uniqueness constraint for API key name and make it optional {pull}47549[#47549] (issue: {issue}46646[#46646]) @@ -383,7 +402,6 @@ Task Management:: Transform:: * Do not fail checkpoint creation due to global checkpoint mismatch {pull}48423[#48423] (issue: {issue}48379[#48379]) * Prevent assignment if any node is older than 7.4 {pull}48055[#48055] (issue: {issue}48019[#48019]) -* Prevent assignment to nodes older than 7.4 {pull}48044[#48044] (issue: {issue}48019[#48019]) * Fix bwc serialization with 7.3 {pull}48021[#48021] * Signal listener early on task _stop failure {pull}47954[#47954] * Use field_caps API for mapping deduction {pull}46703[#46703] (issue: {issue}46694[#46694])