diff options
Diffstat (limited to 'doc/user/project/ml/experiment_tracking')
-rw-r--r-- | doc/user/project/ml/experiment_tracking/img/candidate_v15_11.png | bin | 0 -> 24678 bytes | |||
-rw-r--r-- | doc/user/project/ml/experiment_tracking/img/candidate_v15_7.png | bin | 35164 -> 0 bytes | |||
-rw-r--r-- | doc/user/project/ml/experiment_tracking/img/candidates_v15_11.png | bin | 0 -> 57878 bytes | |||
-rw-r--r-- | doc/user/project/ml/experiment_tracking/img/candidates_v15_7.png | bin | 47800 -> 0 bytes | |||
-rw-r--r-- | doc/user/project/ml/experiment_tracking/img/experiments_v15_11.png | bin | 0 -> 24811 bytes | |||
-rw-r--r-- | doc/user/project/ml/experiment_tracking/img/experiments_v15_7.png | bin | 23475 -> 0 bytes | |||
-rw-r--r-- | doc/user/project/ml/experiment_tracking/index.md | 95 |
7 files changed, 51 insertions, 44 deletions
diff --git a/doc/user/project/ml/experiment_tracking/img/candidate_v15_11.png b/doc/user/project/ml/experiment_tracking/img/candidate_v15_11.png Binary files differnew file mode 100644 index 00000000000..50bcd1e8479 --- /dev/null +++ b/doc/user/project/ml/experiment_tracking/img/candidate_v15_11.png diff --git a/doc/user/project/ml/experiment_tracking/img/candidate_v15_7.png b/doc/user/project/ml/experiment_tracking/img/candidate_v15_7.png Binary files differdeleted file mode 100644 index fb2e2e706d6..00000000000 --- a/doc/user/project/ml/experiment_tracking/img/candidate_v15_7.png +++ /dev/null diff --git a/doc/user/project/ml/experiment_tracking/img/candidates_v15_11.png b/doc/user/project/ml/experiment_tracking/img/candidates_v15_11.png Binary files differnew file mode 100644 index 00000000000..3a2944c92ae --- /dev/null +++ b/doc/user/project/ml/experiment_tracking/img/candidates_v15_11.png diff --git a/doc/user/project/ml/experiment_tracking/img/candidates_v15_7.png b/doc/user/project/ml/experiment_tracking/img/candidates_v15_7.png Binary files differdeleted file mode 100644 index 58dfe94a108..00000000000 --- a/doc/user/project/ml/experiment_tracking/img/candidates_v15_7.png +++ /dev/null diff --git a/doc/user/project/ml/experiment_tracking/img/experiments_v15_11.png b/doc/user/project/ml/experiment_tracking/img/experiments_v15_11.png Binary files differnew file mode 100644 index 00000000000..f7e25c1a0f1 --- /dev/null +++ b/doc/user/project/ml/experiment_tracking/img/experiments_v15_11.png diff --git a/doc/user/project/ml/experiment_tracking/img/experiments_v15_7.png b/doc/user/project/ml/experiment_tracking/img/experiments_v15_7.png Binary files differdeleted file mode 100644 index a7d4a3e559f..00000000000 --- a/doc/user/project/ml/experiment_tracking/img/experiments_v15_7.png +++ /dev/null diff --git a/doc/user/project/ml/experiment_tracking/index.md b/doc/user/project/ml/experiment_tracking/index.md index a7096d633a0..a2c2ab0cc40 100644 --- a/doc/user/project/ml/experiment_tracking/index.md +++ b/doc/user/project/ml/experiment_tracking/index.md @@ -4,78 +4,85 @@ group: Incubation info: Machine Learning Experiment Tracking is a GitLab Incubation Engineering program. No technical writer assigned to this group. --- -# Machine Learning Experiment Tracking **(FREE)** +# Machine learning model experiments **(FREE)** -DISCLAIMER: -Machine Learning Experiment Tracking is an experimental feature being developed by the Incubation Engineering Department, -and will receive significant changes over time. This feature is being release with the aim of getting user feedback, but -is not stable and can lead to performance degradation. See below on how to disable this feature. +FLAG: +On self-managed GitLab, model experiment tracking is disabled by default. +To enable the feature, ask an administrator to [enable the feature flag](../../../../administration/feature_flags.md) named `ml_experiment_tracking`. +On GitLab.com, this feature is in private testing only. -When creating machine learning models, data scientists often experiment with different parameters, configurations, feature -engineering, and so on, to improve the performance of the model. Keeping track of all this metadata and the associated +NOTE: +Model experiment tracking is an [experimental feature](../../../../policy/alpha-beta-support.md). Refer to <https://gitlab.com/gitlab-org/gitlab/-/issues/381660> for feedback and feature requests. + +When creating machine learning models, data scientists often experiment with different parameters, configurations, and feature +engineering to improve the performance of the model. Keeping track of all this metadata and the associated artifacts so that the data scientist can later replicate the experiment is not trivial. Machine learning experiment tracking enables them to log parameters, metrics, and artifacts directly into GitLab, giving easy access later on. -![List of Experiments](img/experiments_v15_7.png) +These features have been proposed: -![Experiment Candidates](img/candidates_v15_7.png) +- Searching experiments. +- Visual comparison of candidates. +- Creating, deleting, and updating candidates through the GitLab UI. -![Candidate Detail](img/candidate_v15_7.png) +For feature requests, see [epic 9341](https://gitlab.com/groups/gitlab-org/-/epics/9341). ## What is an experiment? -An experiment is a collection of comparable model candidates. Experiments can be long lived (for example, when they represent -a use case), or short lived (results from hyperparameter tuning triggered by a merge request), but usually hold model candidates -that have a similar set of parameters and metrics. +In a project, an experiment is a collection of comparable model candidates. +Experiments can be long-lived (for example, when they represent a use case), or +short-lived (results from hyperparameter tuning triggered by a merge request), +but usually hold model candidates that have a similar set of parameters measured +by the same metrics. + +![List of Experiments](img/experiments_v15_11.png) ## Model candidate A model candidate is a variation of the training of a machine learning model, that can be eventually promoted to a version -of the model. The goal of a data scientist is to find the model candidate whose parameter values lead to the best model +of the model. + +![Experiment Candidates](img/candidates_v15_11.png) + +The goal of a data scientist is to find the model candidate whose parameter values lead to the best model performance, as indicated by the given metrics. -Example parameters: +![Candidate Detail](img/candidate_v15_11.png) + +Some example parameters: -- Algorithm (linear regression, decision tree, and so on). +- Algorithm (such as linear regression or decision tree). - Hyperparameters for the algorithm (learning rate, tree depth, number of epochs). - Features included. -## Usage +## Track new experiments and candidates -### User access management +Experiment and trials can only be tracked through the +[MLflow](https://www.mlflow.org/docs/latest/tracking.html) client integration. +See [MLflow client integration](../../integrations/mlflow_client.md) for more information +on how to use GitLab as a backend for the MLflow Client. -An experiment is always associated to a project. Only users with access to the project an experiment is associated with -can view that experiment data. +## Explore model candidates -### Tracking new experiments and trials +Prerequisites: -Experiment and trials can only be tracked through the [MLFlow](https://www.mlflow.org/docs/latest/tracking.html) client -integration. More information on how to use GitLab as a backend for MLFlow Client can be found [at the documentation page](../../integrations/mlflow_client.md). +- You must have at least the Developer role to view experiment data. -### Exploring model candidates +To list the current active experiments, either go to `https/-/ml/experiments` or: -To list the current active experiments, navigate to `https/-/ml/experiments`. To display all trials -that have been logged, along with their metrics and parameters, select an experiment. To display details for a candidate, -select **Details**. +1. On the top bar, select **Main menu > Projects** and find your project. +1. On the left sidebar, select **Packages & registries > Model experiments**. +1. To display all candidates that have been logged, along with their metrics, parameters, and metadata, select an experiment. +1. To display details for a candidate, select **Details**. -### Logging artifacts +## View log artifacts Trial artifacts are saved as [generic packages](../../../packages/generic_packages/index.md), and follow all their -conventions. After an artifact is logged for a candidate, all artifacts logged for the candidate are listed in the -package registry. The package name for a candidate is `ml_candidate_<candidate_id>`, with version `-`. The link to the -artifacts can also be accessed from the **Experiment Candidates** list or **Candidate detail**. - -### Limitations and future - -- Searching experiments, searching trials, visual comparison of trials, and creating, deleting and updating experiments and trials through GitLab UI is under development. - -## Disabling or enabling the Feature - -On self-managed GitLab, ML Experiment Tracking is disabled by default. To enable the feature, ask an administrator to [disable the feature flag](../../../../administration/feature_flags.md) named `ml_experiment_tracking`. -On GitLab.com, this feature is currently on private testing. - -## Feedback, roadmap and reports +limitations. After an artifact is logged for a candidate, all artifacts logged for the candidate are listed in the +package registry. The package name for a candidate is `ml_experiment_<experiment_id>`, where the version is the candidate +IID. The link to the artifacts can also be accessed from the **Experiment Candidates** list or **Candidate detail**. -For updates on the development, refer to the [development epic](https://gitlab.com/groups/gitlab-org/-/epics/8560). +## Related topics -For feedback, bug reports and feature requests, refer to the [feedback issue](https://gitlab.com/gitlab-org/gitlab/-/issues/381660). +- Development details in [epic 8560](https://gitlab.com/groups/gitlab-org/-/epics/8560). +- Add feedback in [issue 381660](https://gitlab.com/gitlab-org/gitlab/-/issues/381660). |