Skip to content

CHANGELOG.md

Latest commit

 

History

History
745 lines (425 loc) · 50.8 KB

CHANGELOG.md

File metadata and controls

745 lines (425 loc) · 50.8 KB
Raw

Changelog

Unreleased

2.0.3 - 2026-05-10

Fixed

  • 🐛 Add PipelineML.__add__ and PipelineML.__radd__ to support pipeline_ml_factory alongside the new feature ability which enables to pass several pipeline_names argument in CLI introduced in kedro==1.2.0 (#690)

2.0.2 - 2026-02-16

Fixed

  • 🐛 Support for pipeline_names argument in CLI added in kedro==1.2.0 (#686)

2.0.1 - 2026-01-18

Fixed

  • 🐛 Enable PipelineML logging with either custom model name following MLflow v3 convention, or custom artifact_path for back-compatibility with older MLflow versions. (#687)

2.0.0 - 2025-11-25

Added

  • ✨ Add support for mlflow>=3.0.0 (#657)

Changed

  • 💥 📌 Drop support for mlflow>=2.0.0,<3.0.0 (#678)
  • 💥 Drop run_id argument of MlflowModelTrackingDataset. In mlflow 3, the model is no longer necessarily tied to a run. It is now possible to specify load_args={"model_uri":"models/<name>} to load a specific model outside the current run. (#678)
  • 💥 Drop artifact_path argument of MlflowModelTrackingDataset. In mlflow 3, this argument is renamed nameand can be passed to the dataset withsave_args={"name":"my_model_name"} (#678)

1.0.2 - 2025-09-27

Fixed

  • 🐛 Treat "databricks-uc" as a special value for mlflow_registry_uri (#675)

1.0.1 - 2025-08-13

Fixed

  • 🐛 Fix run_id logging error while loading MlflowModelRegistryDataset (#671)

1.0.0 - 2025-07-27

Added

  • ✨ Add support for kedro>=1.0.0 (#658)

Changed

  • 💥 📌 Drop support for kedro==0.19.X (#658)

0.14.5 - 2025-07-27

Added

  • 📝 Document how to make partitioned datasets log only outputs of the current run (#652)

Changed

  • 🔇 Update thread-safe related logs to DEBUG level to reduce logging (#650)

Fixed

  • 🐛 Fix KedroPipelineModel.load_context loading from local files with uri filepath for mlflow>=2.21,<3 #656

0.14.4 - 2025-02-20

Added

  • ✨ Add the tracking.disable_tracking.disable_autologging configuration option in mlflow.yml to disable autologging by default. This simplify the workflow for Databricks users who have autologging activated by default, which conflicts with kedro-mlflow ([#610]).
  • ✨ Add tracking.experiment.create_experiment_kwargs.artifact_location and tracking.experiment.create_experiment_kwargs.tags configuration options in mlflow.yml to enable advanced configuration of mlflow experiment created at runtime by kedro-mlflow ([#557]).

Fixed

  • 🐛 Fix type annotations introduced in #633 which are not compatible with python==3.9.

0.14.3 - 2025-02-17

Added

  • ✨ All datasets can now take a metadata parameter (#625, #633)

Fixed

  • 🐛 Reopen the mlflow run before each node to bypass mlflow thread safety and ensure all tracking is done within the same run_id (#623, #624)

0.14.2 - 2025-02-16

Fixed

  • 🐛 👷 Ensure mlflow.yml is package with the package distribution so that kedro mlflow init can find it ([#635]).

0.14.1 - 2025-02-12

Changed

Fixed

  • 🐛 Fix a regression introduced by 0.13.4 and enable again to load an artifact from a different run_id (#622).

0.14.0 - 2025-01-28

Added

  • ✨ Enable to override parameters and the runner at predict time in KedroPipelineModel (#445, #612)

Changed

  • 💥 📌 Pin mlflow>=2.7.0 to support predict parameters for custom models (see above feature)

0.13.4 - 2024-12-14

Fixed

  • 🐛 🚑 Ensure MlflowArtifactDataset logs in the same run that parameters to when using mlflow>=2.18 in combination with ThreadRunner (#613)

0.13.3 - 2024-10-29

Added

  • ✨ Implement missing PipelineML filtering functionalities to let kedro display resume hints and avoid breaking kedro-viz (#377, #601, Calychas)
  • ✨ Sanitize parameters name with unsupported characters to avoid mlflow errors when logging (#595, pascalwhoop)
  • 🔊 Add logs about the exact run_id loaded within a MlflowRegistryDataset because some URI are confusing (e.g. latest) and hard to debug (#552)

Changed

0.13.2 - 2024-10-15

Fixed

  • 🐛 🚑 Fix MlflowArtifactDataset load and save methods to make them compatible with modern datasets without private _load and _save introduced in kedro-datasets>=5.0.0 (#598)

0.13.1 - 2024-09-24

Added

Fixed

  • 🐛 Refactor MlflowMetricsHistoryDataset to avoid using get_all_metrics internally because this function cannot save metrics on a remote server (#582)

0.13.0 - 2024-09-01

Added

  • ✨ Add support for loading model with alias in MlflowModelRegistryDataset (#553)

Changed

  • 💥 📌 Officially drop support for mlflow<1.29.0 which was implicit since the introduction of km.random_name resolver in #481 (#571)

  • 📝 🔊 Various improvements to documentation and logging (#549, #554, #567)

Fixed

  • 🐛 Implement only_nodes_with_namespace and __sub__ methods in PipelineML for kedro-viz>=9.2.0 compatibility (#569)

0.12.2 - 2024-04-18

Added

  • 👷 📦 Add build distribution (instead of source only distribution) to PyPI release to make install easier and faster (#515)
  • 📝 Document the ability to update configuration at runtime (#395)

Changed

  • ✨ Project level CLI commands kedro mlflow init, kedro mlflow ui and kedro mlflow modelify now work even inside a subdirectory and not at the root of the kedro project to be consistent with kedro>0.19.4 (#531)

Fixed

  • 🐛 Proxy import of private kedro functions _is_project and _find_kedro_project to be resilient to changes (#531)

0.12.1 - 2024-02-09

Added

  • ✨ Add a km.random_name resolver which enables to use auto-generated names for kedro runs instead of pipeline name in the mlflow.yml configuration file (#426)

Fixed

0.12.0 - 2023-12-19

Added

  • ✨ Add support for python 3.11 (#450, rxm7706)
  • ✨ ⬆️ Add support for pydantic v2 (#476)
  • ✨ ⬆️ Add support for kedro==0.19.X (#)

Changed

  • 💥 :⬆️: Drop support for kedro==0.18.X series.
  • 💥 ✨ Change default copy_mode to "assign" in KedroPipelineModel because this is the most efficient setup (and usually the desired one) when serving a Kedro Pipeline as a Mlflow model. This is different from Kedro's default which is to deepcopy the dataset (#463).
  • 💥 ♻️ MlflowArtifactDataset.__init__ method data_set argument is renamed dataset to match new Kedro conventions (#391).
  • 💥 ♻️ Rename the following DataSets with the Dataset suffix (without capitalized S) to match new kedro conventions from kedro>=0.19 and onwards (#439, ShubhamZoro):
    • MlflowArtifactDataSet->MlflowArtifactDataset
    • MlflowAbstractModelDataSet->MlflowAbstractModelDataset
    • MlflowModelRegistryDataSet->MlflowModelRegistryDataset
    • MlflowMetricDataSet->MlflowMetricDataset
    • MlflowMetricHistoryDataSet->MlflowMetricHistoryDataset
  • 💥 ♻️ Rename the following DataSets to make their use more explicit, and use the Dataset suffix:
    • MlflowModelLoggerDataSet->MlflowModelTrackingDataset (#391)
    • MlflowModelSaverDataSet->MlflowModelLocalFileSystemDataset (#391)
    • MlflowMetricsDataSet->MlflowMetricsHistoryDataset (#440)

0.11.10 - 2023-10-03

Fixed

  • 🐛 Avoid error when using kedro==0.18.1 with TemplatedConfigLoader and no mlflow.yml configuration file (#452, sami-sweng)

0.11.9 - 2023-07-23

Fixed

  • 🐛 Make kedro-mlflow hook log parameters when the project is configured with the OmegaConfigLoader instead of raising an error (#430)

Removed

0.11.8 - 2023-02-13

Added

  • ✨ Added support for Mlflow 2.0 (#390)

  • ✨ The modelify command now accepts a --run-name to specifiy the run name where the model is logged (#408)

Fixed

  • 📝 Update incorrect documentation about model registry with local relative filepath (#400)

  • 🐛 The modelify command now creates a conda environment based on your environment python and kedro versions instead of hardcoded python=3.7 and kedro=0.16.5 (#405)

  • 🐛 The modelify command now uses correctly the --pip-requirements argument instead of raising an error (#405)

  • 🐛 The modelify command now uses modelify as a default run name (#408)

0.11.7 - 2023-01-28

Added

  • ✨ Added a MlflowModelRegistryDataSet in kedro_mlflow.io.models to enable fetching a mlflow model from the mlflow model registry by its name(#260)

Fixed

  • 🐛 Use __default__ as a run name if the pipeline is not specified in the kedro run commmand to avoid empty names (#392)

0.11.6 - 2023-01-09

Changed

  • ✨ kedro-mlflow now uses the default configuration (ignoring mlflow.yml) if an active run already exists in the process where the pipeline is started, and uses this active run for logging. This enables using kedro-mlflow with an orchestrator which starts mlflow itself before running kedro (e.g. airflow, the mlflow run command, AzureML...) (#358)

0.11.5 - 2022-12-12

Added

  • ✨ Added an extra server.mlflow_registry_uri key in mlflow.yml to set the mlflow registry uri. (#260)
  • ✨ Add support for authorization with expiring tokens by adding an extra server.request_header_provider entry in mlflow.yml (#357)

Fixed

  • 🐛 MlflowArtifactDataSet.load() now correctly loads the artifact when both artifact_path and run_id arguments are specified. Previous fix in 0.11.4 did not work because when the file already exist locally, mlflow did not download it again so tests were incorrectly passing (#362)

Removed

  • 🔥 💥 Remove reload_kedro_mlflow line magic for notebook because kedro will deprecate the entrypoint in 0.18.3. It is still possible to access the mlflow client associated to the configuration in a notebook with context.mlflow.server._mlflow_client (#349). This is not considered as a breaking change since apparently no one uses it according to a discussion with kedro's team.

0.11.4 - 2022-10-04

Fixed

  • 🐛 MlflowArtifactDataSet.load() now correctly loads the artifact when both artifact_path and run_id arguments are specified instead of raising an error (#362)

0.11.3 - 2022-09-06

Changed

  • 🔊 kedro-mlflow has its default logging level set to INFO. This was the default for kedro<=0.18.1. For kedro>=0.18.2, you can change the level in logging.yml (#348)

Fixed

  • 🐛 kedro-mlflow now uses the package_name as experiment name by default if it is not specified. This is done to ensure consistency with the behaviour with no mlflow.yml file (#328)
  • 📝 Update broken links to the most recent kedro and mlflow documentation

0.11.2 - 2022-08-28

Changed

  • ✨ kedro-mlflow now runs even without a mlflow.yml file in your conf/<env> folder. As a consequence, running kedro mlflow init is now optional and should be only used for advanced configuration. (#328)

0.11.1 - 2022-07-06

Fixed

  • 🐛 Make pipeline_ml_factory now correctly uses kpm_kwargs and log_model_kwargs instead of always using the default values. (#329)
  • 🐛 kedro mlflow init command no longer raises both a success and an error message when the command is failing. (#336)

Changed

  • ♻️ Refactor KedroMlflowConfig which no longer needs the project_path at instantiation. The uri validaiton is done at setup() time to be able to use the configuration not at a root of a kedro project. This is not considered as a breaking change, because the recommended way to retrieve the config is to use session.load_context().mlflow which automatically calls setup() and hence behaviour inside a kedro project is unmodified. (#314)

0.11.0 - 2022-06-18

Added

  • ✨ 💥 The MLFLOW_TRACKING_URI environment variable is now used as the default tracking uri if the server.mlflow_tracking_uri config key is None. The mlflow.yml is changed to server: mlflow_tracking_uri: null to enforce this new behaviour as the default value. If the environment variable does not exists, it will behave like before. (#321).

Changed

  • ♻️ 💥 Unify the MlflowPipelineHook and MlflowNodeHook in a single MlflowHook to ensure consistency in registration order (#315)
  • ♻️ 🧑‍💻 💥 The get_mlflow_config public function is removed. If you need to access the mlflow configuration, you can do it automatically in the context mlflow attribute, e.g. session.load_context().mlflow (#310)

Removed

  • ⚰️ 💥 Remove unused stores_environment_variables configuration option. This key must be removed from mlflow.yml.
  • ⬆️ 🐛 Upgrade requirements to make support for kedro>=0.18.1, kedro<0.19.0 explicit. This is the only valid compatibility range since kedro-mlflow==0.10.0, but requirements had not been updated yet (#309).

0.10.0 - 2022-05-15

Added

  • ⬆️ Add support for kedro==0.18.1 which was broken due to kedro's removal of _active_session private global variable (#309).

Fixed

  • 📝 Fix typo in documentation (#302)

Changed

  • ♻️ 💥 Refactor the get_mlflow_config function which now takes context instead of session as input (#309)

Removed

  • 💥 ⬇️ Drop support for kedro=0.18.0. kedro-mlflow now supports only kedro>=0.18.1, kedro<0.19.0 (#309).

0.9.0 - 2022-04-01

Added

  • ✨ Add support for kedro=0.18.X (#290)
  • ✨ kedro-mlflow is now available on conda-forge and can be installed with conda install kedro-mlflow. This is retroactive to kedro-mlflow==0.8.1 (#118)

Removed

  • 💥 🗑️ Drop support for kedro=0.17.X (#290)

0.8.1 - 2022-02-13

Added

  • ✨ Open the UI in the default browser when the mlflow_tracking_uri in mlflow.yml is a http address instead of launching the ui server. (#275)

Fixed

  • 🐛 Force the input dataset in KedroPipelineModel to be a MemoryDataset to remove unnecessary dependency to the underlying Kedro AbstractDataset used during training (#273)
  • 🐛 Make MlflowArtifactDataset correctly log in mlflow Kedro DataSets without a _path attribute like kedro.io.PartitionedDataset (#258).
  • 🐛 Automatically persist pipeline parameters when calling the kedro mlflow modelify command for consistency with how PipelineML objects are handled and for ease of use (#282).

0.8.0 - 2022-01-05

Added

  • ✨ Add a kedro mlflow modelify command to export a pipeline as a mlflow model (#261)
  • 📝 Format code blocks in documentation with blacken-docs
  • 👷 Enforce the use of black and isort in the CI to enforce style guidelines for developers

Changed

  • ✨ 💥 The pipeline_ml_factory accepts 2 new arguments log_model_kwargs (which will be passed as is to mlflow.pyfunc.log_model) and kpm_kwargs (which will be passed as is to KedroPipelineModel). This ensures perfect consistency with mlflow API and offers new possibility like saving the project source code alongside the model (#67). Note that model_signature, conda_env and model_name arguments are removed, and replace respectively by log_model_kwargs["signature"], log_model_kwargs["conda_env"] and log_model_kwargs["artifact_path"].
  • ✨ 💥 The KedroPipelineModel custom mlflow model now accepts any kedro Pipeline as input (provided they have a single DataFrame input and a single output because this is an mlflow limitation) instead of only PipelineML objects. This simplifies the API for user who want to customise the model logging (#171). KedroPipelineModel.__init__ argument pipeline_ml is renamed pipeline to reflect this change.
  • 🗑️ kedro_mlflow.io.metrics.MlflowMetricsDataSet is no longer deprecated because there is no alternative for now to log many metrics at the same time.
  • 💥 Refactor mlflow.yml to match mlflow's API (#77). To migrate projects with kedro<0.8.0, please update their mlflow.yml with kedro mlflow init --force command.

Fixed

  • 🐛 KedroMlflowConfig.setup() methods now sets the experiment globally to ensure all runs are launched under the experiment specified in the configuration even in interactive mode (#256).

Removed

  • 🔥 💥 KedroMlflowConfig and get_mlflow_config were deprecated since 0.7.3 and are now removed from kedro_mlflow.framework.context. Direct import must now use kedro_mlflow.config.

0.7.6 - 2021-10-08

Fixed

  • 🐛 The reserved keyword "databricks" is no longer converted to a local filepath before setting the MLFLOW_TRACKING_URI to enable integration with databricks managed platform. (#248)

0.7.5 - 2021-09-21

Added

  • ✨ Add support for notebook use. When a notebook is opened via a kedro command (e.g. kedro jupyter notebook), you can call the %reload_kedro_mlflow line magic to setup mlflow configuration automatically. A mlflow_client to the database is also created available as a global variable (#124).
  • 📝 Add automatic API documentation through docstrings for better consistency between code and docs (#110). All docstrings are not updated yet and it will be a long term work.

Changed

  • ♻️ KedroMlflowConfig was refactored with pydantic for improved type checking when loading configuration, overall robustness and autocompletion. Its keys have changed, but it is not considered as a user facing changes since the public function get_mlflow_config() and KedroMlflowConfig().setup() are not modified.

  • 🗑️ The kedro.framework.context folder is moved to kedro.config for consistency with the Kedro repo structure: get_mlflow_config import must change from from kedro_mlflow.framework.context import get_mlflow_config to from kedro_mlflow.config import get_mlflow_config.

0.7.4 - 2021-08-30

Added

  • ✨ Create an MlflowMetricDataSet to simplify the existing metric API. It enables logging a single float as a metric, eventually automatically increasing the "step" if the metric is going to be updated during time (#73)
  • ✨ Create an MlflowMetricHistoryDataSet to simplify the existing metric API. It enables logging the evolution of a given metric during training. (#73)

Fixed

  • 🐛 Dictionnary parameters with integer keys are now properly logged in mlflow when flatten_dict_params is set to True in the mlflow.yml instead of raising a TypeError (#224)
  • 🐛 The user defined sep parameter of the mlflow.yml (defined in node section) is now used even if the parameters dictionnary has a depth>=2 (#230)

Changed

  • ♻️ Move flatten_dict function to hooks.utils folder and rename it _flatten_dict to make more explicit that it is not a user facing function which should not be used directly and comes with no guarantee. This is not considered as a breaking change since it is an undocumented function.
  • 🗑️ Deprecate MlflowMetricsDataSet in favor of the 2 new datasets MlflowMetricDataSet and MlflowMetricHistoryDataSet newly added. It will be removed in kedro-mlflow==0.8.0.

0.7.3 - 2021-08-16

Added

  • ✨ Update the MlflowArtifactDataSet.load() method to download the data from the run_id if it is specified instead of using the local filepath. This can be used for instance to continue training from a pretrained model or to retrieve the best model from an hyperparameter search (#95)

0.7.2 - 2021-05-02

Fixed

  • 🐛 Remove global CLI command new (which was not implemented yet) to make project CLI commands available. It is not possible to have 2 CLI groups (one at global level , one at project level) because of a bug in kedro==0.17.3 (#193)

0.7.1 - 2021-04-09

Added

  • ✨ It is now possible to deactivate tracking (for parameters and datasets) by specifying a key disabled_tracking: pipelines: [<pipeline-name>] in the mlflow.yml configuration file. (#92)

  • ✨ The kedro mlflow ui command host and port keys can be overwritten at runtime (#187)

Fixed

  • 🐛 The kedro mlflow ui command now reads properly the ui:host and ui:port keys from the mlflow.yml which were incorrectly ignored (#187)

0.7.0 - 2021-03-17

Added

  • ⬆️ kedro-mlflow now supports kedro>=0.17.1 (#144).

Changed

  • 📌 Drop support for kedro==0.17.0, since the kedro core team made a breaking change in 0.17.1. All future plugin updates will be only compatible with kedro>=0.17.1.

0.6.0 - 2021-03-14

Added

Fixed

  • 🐛 The support of kedro==0.17.0 automatically makes the CLI commands available when the configuration is declared in a pyproject.toml instead of a .kedro.yml, which was not the case in previous version despite we claim it was (#157).

Changed

  • 📌 Drop support for kedro==0.16.x. All future plugin updates will be only compatible with kedro>=0.17.0.

0.5.0 - 2021-02-21

Added

  • ✨ A new long_parameters_strategy key is added in the mlflow.yml (under in the hook/node section). You can specify different strategies (fail, truncate or tag) to handle parameters over 250 characters which cause crashes for some mlflow backend. (#69)
  • ✨ Add an env parameter to kedro mlflow init command to specify under which conf/ subfolder the mlflow.yml should be created. (#159)
  • ✨ The input parameters of the inference pipeline of a PipelineML object are now automatically pickle-ised and converted as artifacts. (#158)
  • 📝 Detailed documentation on how to use pipeline_ml_factory function, and more generally on how to use kedro-mlflow as mlops framework. This comes from an example repo kedro-mlflow-tutorial. (#16)

Fixed

  • 📌 Pin the kedro version to force it to be strictly inferior to 0.17 which is not compatible with current kedro-mlflow version (#143)
  • ✨ It is no longer assumed for the project to run that the mlflow.yml is located under conf/base. The project will run as soon as the configuration file is discovered by the registered ConfigLoader (#159)

Changed

  • ⚡ 💥 The KedroPipelineModel.load_context() method now loads all the DataSets in memory in the DataCatalog. It is also now possible to specify the runner to execute the model as well as the copy_mode when executing the inference pipeline (instead of deepcopying the datasets between each nodes which is kedro's default). This makes the API serving with mlflow serve command considerably faster (~20 times faster) for models which need compiling (e.g. keras, tensorflow ...) (#133)
  • ✨ The CLI projects commands are now always accessible even if you have not called kedro mlflow init yet to create a mlflow.yml configuration file (#159)

0.4.1 - 2020-12-03

Added

  • ✨ It is now possible to supply credentials for the mlflow tracking server within mlflow.yml and credentials.yml. They will be exported as environment variables during the run. (#31)

Fixed

  • 🐛 Fix TypeError: unsupported operand type(s) for /: 'str' and 'str' when using MlflowArtifactDataSet with MlflowModelLocalFileSystemDataset (#116)
  • 📝 Fix various docs typo (#6)
  • 🐛 When the underlying Kedro pipeline fails, the associated mlflow run is now marked as 'FAILED' instead of 'FINISHED'. It is rendered with a red cross instead of the green tick in the mlflow user interface (#121).
  • 🐛 Fix a bug which made KedroPipelineModel impossible to load if one of its artifact was a MlflowModel<Saver/Logger>DataSet. These datasets were not deepcopiable because of one their attributes was a module (#122).

Changed

  • 📝 Refactor doc structure for readability (#6)
  • ⚡ 💥 The KedroMlflowConfig no longer creates the mlflow experiment and access to the mlflow tracking server when it is instantiated. A new setup() method sets up the mlflow configuration (tracking uri, credentials and experiment management) but must now be called explicitly. (#97)

0.4.0 - 2020-11-03

Added

  • ⬆️ kedro-mlflow now supports kedro>=0.16.5 (#62)
  • ✨ kedro-mlflow now supports configuring the project in pyproject.toml (Only for kedro>=0.16.5) (#96)
  • ✨ pipeline_ml_factory now accepts that inference pipeline inputs may be in training pipeline inputs (#71)
  • ✨ pipeline_ml_factory now infer automatically the schema of the input dataset to validate data automatically at inference time. The output schema can be declared manually in model_signature argument (#70)
  • ✨ Add two DataSets for model logging and saving: MlflowModelTrackingDataset and MlflowModelLocalFileSystemDataset (#12)
  • ✨ MlflowPipelineHook and MlflowNodeHook are now auto-registered if you use kedro>=0.16.4 (#29)

Fixed

  • ⚡ get_mlflow_config now uses the Kedro ProjectContext ConfigLoader to get configs (#66). This indirectly solves the following issues:
    • get_mlflow_config now works in interactive mode if load_context is called with a path different from the working directory (#30)
    • kedro_mlflow now works fine with kedro jupyter notebook independently of the working directory (#64)
    • You can use global variables in mlflow.yml which is now properly parsed if you use a TemplatedConfigLoader (#72)
  • 🐛 MlflowMetricsHistoryDataset now saves in the specified run_id instead of the current one when the prefix is not specified (#62)
  • 📝 Other bug fixes and documentation improvements (#6, #99)

Changed

  • ✨ 💥 The KedroPipelineModel now unpacks the result of the inference pipeline and no longer returns a dictionary with the name in the DataCatalog but only the predicted value (#93)
  • ♻️ 💥 The PipelineML.extract_pipeline_catalog is renamed PipelineML._extract_pipeline_catalog to indicate it is a private method and is not intended to be used directly by end users who should rely on PipelineML.extract_pipeline_artifacts (#100)
  • 🏗️ 💥 The MlflowArtifactDataSet is moved from kedro_mlflow.io folder to kedro_mlflow.io.artifacts. (#109)
  • 🏗️ 💥 The MlflowMetricsDataSet is moved from kedro_mlflow.io folder to kedro_mlflow.io.metrics. (#109)

Removed

  • ♻️ 💥 kedro mlflow init command is no longer declaring hooks in run.py. You must now register your hooks manually in the run.py if you use kedro>=0.16.0, <0.16.3 (#62).
  • 🔥 Remove pipeline_ml function which was deprecated in 0.3.0. It is now replaced by pipeline_ml_factory (#105)
  • 🔥 Remove MlflowDataSet dataset which was deprecated in 0.3.0. It is now replaced by MlflowArtifactDataSet (#105)

0.3.0 - 2020-10-11

Added

  • ✨ Add dataset MlflowMetricsDataSet for metrics logging (#9) and update documentation for metrics.
  • 👷 Add CI workflow create-release to ensure release consistency and up-to-date CHANGELOG. (#57, #68)
  • 📝 Add templates for issues and pull requests (#57, #68)

Fixed

  • 🐛 Versioned datasets artifacts logging are handled correctly (#41)
  • 🐛 MlflowDataSet handles correctly datasets which are inherited from AbstractDataset (#45)
  • ⚡ Change the test in _generate_kedro_command to accept both empty Iterables(default in CLI mode) and None values (default in interactive mode) (#50)
  • ⚡ Force to close all mlflow runs when a pipeline fails. It prevents further execution of the pipeline to be logged within the same mlflow run_id as the failing pipeline. (#10)
  • 📝 Fix various documentation typos (#34, #35, #36 and more)
  • 📝 Update README (add badges for readibility, add a "main contributors" section to give credit, fix typo in install command, link to milestones for more up-to-date priorities) (#57, #68)
  • 👷 Fix bug in CI deployment workflow and rename it to publish (#57, #68)
  • 🐛 Fix a bug in MlflowDataSet which sometimes failed to log on remote storage (S3, Azure Blob storage) with underlying log_artifacts when the kedro's AbstractDataset._filepath was a pathlib.PurePosixPath object instead of a string (#74).
  • 👷 Add a CI for release candidate creation and use actions to enforce semantic versioning and Keep a Changelog format.

Changed

  • ♻️ Remove conda_env and model_name arguments from MlflowPipelineHook and add them to PipelineML and pipeline_ml. This is necessary for incoming hook auto-discovery in future release and it enables having multiple PipelineML in the same project (#58). This mechanically fixes #54 by making conda_env path absolute for airflow support.
  • ♻️ 💥 flatten_dict_params, recursive and sep arguments of the MlflowNodeHook are moved to the mlflow.yml config file to prepare plugin auto registration. This also modifies the run.py template (to remove the args) and the mlflow.yml keys to add a hooks entry. (#59)
  • 👷 Rename CI workflow to test (#57, #68)
  • ⚡ The input_name attributes of PipelineML is now a python property and makes a check at setting time to prevent setting an invalid value. The check ensures that input_name is a valid input of the inference pipeline.

Deprecated

  • 🗑️ Deprecate MlflowDataSet which is renamed as MlflowArtifactDataSet for consistency with the other datasets. It will raise a DeprecationWarning in this realease, and will be totally supressed in next minor release. Please update your catalog.yml entries accordingly as soon as possible. (#63)
  • 🗑️ Deprecate pipeline_ml which is renamed as pipeline_ml_factory to avoid confusion between a PipelineML instance and the helper function to create PipelineMl instances from Kedro Pipelines.

0.2.1 - 2018-08-06

Added

📝 Many documentation improvements:

  • Add a Code of conduct
  • Add a Contributing guide
  • Refactor README.md to separate clearly from documentation
  • Fix broken links
  • Fix bad markdown rendering
  • Split old README.md information in dedicated sections

Changed

  • 🐛 Enable pipeline_ml to accept artifacts (encoder, binarizer...) to be "intermediary" outputs of the pipeline and not only "terminal" outputs (i.e. node outputs which are not re-used as another node input). This closes a bug discovered in a more general discussion in #16.
  • ♻️ Only non-empty CLI arguments and options are logged as tags in MLflow (#32)

0.2.0 - 2020-07-18

Added

  • ✅ Bump the codebase test coverage to 100% to improve stability
  • 🚨 Improve rendering of template with a trailing newline to make them black-valid
  • ✨ Add a PipelineML.extract_pipeline_artifacts methods to make artifacts retrieving easier for a given pipeline
  • 🎉 Use an official kedro release (>0.16.0, <0.17.0) instead of the development branch

Changed

  • 🏗️ 💥 hooks, context and cli folders are moved to framework to fit kedro new folder architecture
  • ♻️ 💥 Rename get_mlflow_conf in get_mlflow_config for consistency (with ConfigLoader, KedroMlflowConfig...)
  • ♻️ 💥 Rename keys of KedroMlflowConfig.to_dict() to remove the "_opts" suffix for consistency with the KedroMlflowConfig.from_dict method

Fixed

  • 🙈 Add debug folder to gitignore for to avoid involuntary data leakage
  • 🐛 Remove the inadequate warning "You have not initialized your project yet" when calling kedro mlflow init
  • 🐛 Remove useless check to see if the commands are called inside a Kedro project since the commands are dynamically displayed based on whether the call is made inside a kedro project or not
  • 🐛 Fix typos in error messages
  • 🐛 Fix hardcoded path to the run.py template
  • 🐛 Make not implemented function raise a NotImplementError instead of failing silently
  • 🐛 Fix wrong parsing when the mlflow_tracking_uri key of the mlflow.yml configuration file was an absolute local path
  • ⚰️ Remove unused KedroMlflowContextClass
  • 🐛 Force the MlflowPipelineHook.before_pipeline_run method to set the mlflow_tracking_uri to the one from the configuration to enforce configuration file to be prevalent on environment variables or current active tracking uri in interactive mode
  • 🐛 Fix wrong environment parsing case when passing a conda environment as a python dictionary in MlflowPipelineHook
  • 🐛 Fix wrong artifact logging of MlflowDataSet when a run was already active and the save method was called in an interactive python session.
  • ♻️ 💥 Force the user to declare an input_name for a PipelineMl object to fix difficult inference of what is the pipeline input
  • ♻️ Update run.py template to fit kedro new one.
  • ♻️ Force _generate_kedro_commands to separate an option and its arguments with a "=" sign for readibility

0.1.0 - 2020-04-18

Added

  • ✨ Add cli kedro mlflow init to udpdate the template and kedro mlflow ui to open mlflow user interface with your project configuration
  • ✨ Add hooks MlflowNodeHook and MlflowPipelineHook for parameters autologging and model autologging
  • ✨ Add MlflowDataSet for artifacts autologging
  • ✨ Add PipelineMl class and its pipeline_ml factory for pipeline packaging and service