You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: articles/machine-learning/azure-machine-learning-release-notes.md
+100Lines changed: 100 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -17,6 +17,106 @@ In this article, learn about Azure Machine Learning releases. For the full SDK
17
17
18
18
See [the list of known issues](resource-known-issues.md) to learn about known bugs and workarounds.
19
19
20
+
## 2020-05-26
21
+
22
+
### Azure Machine Learning SDK for Python v1.MinorVersionTBD.0
23
+
24
+
+**New features**
25
+
+[Insert new features below. Reference articles and/or doc pages]
26
+
27
+
+**Preview features**
28
+
+[Contrib features below]
29
+
30
+
+**Breaking changes**
31
+
+[Reference upcoming breaking changes and old API support drop date]
32
+
33
+
+**Bug fixes and improvements**
34
+
+**azureml-automl-core**
35
+
+ Fixed the bug where a warning may be printed during `get_output` that asked user to downgrade client.
36
+
+ Updated Mac to rely on cudatoolkit=9.0 as it is not available at version 10 yet.
37
+
+ Removing restrictions on phrophet and xgboost models when trained on remote compute.
38
+
+ Improved logging in AutoML
39
+
+ The error handling for custom featurization in forecasting tasks was improved.
40
+
+ Added functionality to allow users to include lagged features to generate forecasts.
41
+
+ Updates to error message to correctly display user error.
42
+
+ Support for cv_split_column_names to be used with training_data
43
+
+ Update logging the exception message and traceback.
44
+
+**azureml-automl-runtime**
45
+
+ Enable guardrails for forecasting missing value imputations.
46
+
+ Improved logging in AutoML
47
+
+ Added fine grained error handling for dataprep exceptions
48
+
+ Removing restrictions on phrophet and xgboost models when trained on remote compute.
49
+
+`azureml-train-automl-runtime` and `azureml-automl-runtime` have updated dependencies for `pytorch`, `scipy`, and `cudatoolkit`. we now support `pytorch==1.4.0`, `scipy>=1.0.0,<=1.3.1`, and `cudatoolkit==10.1.243`.
50
+
+ The error handling for custom featurization in forecasting tasks was improved.
51
+
+ The forecasting data set frequency detection mechanism was improved.
52
+
+ Fixed issue with Prophet model training on some data sets.
53
+
+ The auto detection of max horizon during the forecasting was improved.
54
+
+ Added functionality to allow users to include lagged features to generate forecasts.
55
+
+ Adds functionality in the forecast function to enable providing forecasts beyond the trained horizon without re-training the forecasting model.
56
+
+ Support for cv_split_column_names to be used with training_data
57
+
+**azureml-contrib-automl-dnn-forecasting**
58
+
+ Improved logging in AutoML
59
+
+**azureml-contrib-dataset**
60
+
+ Significantly Improved error text in case of Dataset execution failures.
61
+
+**azureml-contrib-mir**
62
+
+ Added support for Windows services in ManagedInferencing
63
+
+ Remove old MIR workflows such as attach MIR compute, SingleModelMirWebservice class - Clean out model profiling placed in contrib-mir package
64
+
+**azureml-contrib-pipeline-steps**
65
+
+ Quick fix for ParallelRunStep where loading from YAML was broken
66
+
+ ParallelRunStep is released to General Availability - azureml.contrib.pipeline.steps has a deprecation notice and is move to azureml.pipeline.steps - new features include: 1. Datasets as PipelineParameter 2. New parameter run_max_retry 3. Configurable append_row output file name
67
+
+**azureml-contrib-reinforcementlearning**
68
+
+ RL Load testing tool
69
+
+ RL estimator has smart defaults
70
+
+**azureml-core**
71
+
+ Remove old MIR workflows such as attach MIR compute, SingleModelMirWebservice class - Clean out model profiling placed in contrib-mir package
72
+
+ Fixed the information provided to the user in case of profiling failure: included request id and reworded the message to be more meaningful. Added new profiling workflow to profiling runners
73
+
+ Significantly Improved error text in case of Dataset execution failures.
74
+
+ Workspace private link cli support added.
75
+
+ Added an optional parameter `invalid_lines` to `Dataset.Tabular.from_json_lines_files` that allows for specifying how to handle lines that contain invalid JSON.
76
+
+ We will be deprecating the run based creation of compute in the next release. We recommend creating an actual Amlcompute cluster as a persistent compute target, and using the cluster name as the compute target in your run configuration. See example notebook here: aka.ms/amlcomputenb
77
+
+**azureml-dataprep**
78
+
+ Made warning to upgrade pyarrow version more explicit.
79
+
+ Significantly Improved error text in case of Dataset execution failures.
80
+
+ Improved error handling and message returned in case of failure to execute dataflow.
81
+
+ ParallelRunStep is released to General Availability - azureml.contrib.pipeline.steps has a deprecation notice and is move to azureml.pipeline.steps - new features include: 1. Datasets as PipelineParameter 2. New parameter run_max_retry 3. Configurable append_row output file name
82
+
+**azureml-interpret**
83
+
+ Documentation updates to azureml-interpret package.
84
+
+ fixed interpretability packages and notebooks to be compatible with latest sklearn update
85
+
+**azureml-opendatasets**
86
+
+ return None when there is no data returned.
87
+
+ Improve the performance of to_pandas_dataframe.
88
+
+`azureml-train-automl-runtime` and `azureml-automl-runtime` have updated dependencies for `pytorch`, `scipy`, and `cudatoolkit`. we now support `pytorch==1.4.0`, `scipy>=1.0.0,<=1.3.1`, and `cudatoolkit==10.1.243`.
89
+
+**azureml-pipeline-core**
90
+
+ Quick fix for ParallelRunStep where loading from YAML was broken
91
+
+ ParallelRunStep is released to General Availability - azureml.contrib.pipeline.steps has a deprecation notice and is move to azureml.pipeline.steps - new features include: 1. Datasets as PipelineParameter 2. New parameter run_max_retry 3. Configurable append_row output file name
92
+
+**azureml-pipeline-steps**
93
+
+ Deprecated azureml.dprep.Dataflow as a valid type for input data.
94
+
+ Quick fix for ParallelRunStep where loading from YAML was broken
95
+
+ ParallelRunStep is released to General Availability - azureml.contrib.pipeline.steps has a deprecation notice and is move to azureml.pipeline.steps - new features include: 1. Datasets as PipelineParameter 2. New parameter run_max_retry 3. Configurable append_row output file name
96
+
+**azureml-telemetry**
97
+
+ Update logging the exception message and traceback.
98
+
+**azureml-train-automl-client**
99
+
+ Improved logging in AutoML
100
+
+ Updates to error message to correctly display user error.
101
+
+ Support for cv_split_column_names to be used with training_data
102
+
+ Deprecated azureml.dprep.Dataflow as a valid type for input data.
103
+
+ Updated Mac to rely on cudatoolkit=9.0 as it is not available at version 10 yet.
104
+
+ Removing restrictions on phrophet and xgboost models when trained on remote compute.
105
+
+`azureml-train-automl-runtime` and `azureml-automl-runtime` have updated dependencies for `pytorch`, `scipy`, and `cudatoolkit`. we now support `pytorch==1.4.0`, `scipy>=1.0.0,<=1.3.1`, and `cudatoolkit==10.1.243`.
106
+
+ Added functionality to allow users to include lagged features to generate forecasts.
107
+
+**azureml-train-automl-runtime**
108
+
+ Improved logging in AutoML
109
+
+ Added fine grained error handling for dataprep exceptions
110
+
+ Removing restrictions on phrophet and xgboost models when trained on remote compute.
111
+
+`azureml-train-automl-runtime` and `azureml-automl-runtime` have updated dependencies for `pytorch`, `scipy`, and `cudatoolkit`. we now support `pytorch==1.4.0`, `scipy>=1.0.0,<=1.3.1`, and `cudatoolkit==10.1.243`.
112
+
+ Updates to error message to correctly display user error.
113
+
+ Support for cv_split_column_names to be used with training_data
114
+
+**azureml-train-core**
115
+
+ Added a new set of HyperDrive specific exceptions. azureml.train.hyperdrive will now throw detailed exceptions.
0 commit comments