automlops

[1.3.0] - 2024-5-22

Added

Changed

• Refactored code to follow OOP paradigm
• Made kfp a required dependency of AutoMLOps
• kfp v2 code is now generated and required for use with AutoMLOps; before this version, kfp<2.0.0 was required. kfp v2 is not always backwards compatibile with v1
• Removed support for reading/writing kfp components directly. AutoMLOps now only supports our standard AutoMLOps.component and AutoMLOps.pipeline syntax rather than dsl.component
• Removed requirement that input parameters for AutoMLOps.component be python primitives only
• Removed the need for .AutoMLOps-cache; files are no longer written to an itermediary location and stored in memory instead
• Updated examples directory to support new changes/updates

Fixed

[1.2.9] - 2024-1-18

Added

• Added setup_model_monitoring parameter to AutoMLOps.generate and orchestration/configs.py
• Added the creation of a model_monitoring/ directory and create_model_monitoring_job.sh script to generate
• Added a new AutoMLOps.monitor function, along with relevant templates and tests
• Added in a monitoring section into the generated config/defaults.yaml file
• Added logging.googleapis.com to the list of potentially required apis.
• Created 2 new functions for giving warnings when running AutoMLOps.monitor: get_model_monitoring_min_permissions and get_model_monitoring_recommended_roles in utils.py

Changed

• Updated the services/submission_service/main.py.j2 to include elements for automatic retraining based on monitoring anomaly logs, and adding in labels to the submit.
• Updated services/submission_service/requirements.txt to include google-cloud-storage.
• Updated the README.md.j2 to reflect the optional creation of the new model_monitoring/ directory.
• Changed the format for how config/defaults.yaml file gets written; this file is now written using the write_yaml_file function (yaml.safe_dump) in utils.py.
• Updated the account_permissions_warning function in utils.py to include a new operation: operation='model_monitoring'
• Updated the introductory example, readme, and user guide to reflect the new monitoring capabilities.
• Renamed validate_schedule in utils.py to validate_use_ci to reflect new requirements for model monitoring.

Fixed

[1.2.8] - 2024-1-09

Added

• Support for Output annotations with AutoMLOps.component

Changed

Fixed

[1.2.7] - 2023-11-06

Added

Changed

• Updated state_bucket naming schema
• Updated unit tests to reflect state_bucket naming schema
• Updated outputs.tf block to not include csr if use_ci=False

Fixed

[1.2.6] - 2023-10-05

Added

Changed

• Updated default to not disable api services
• Changed required_permissions and recommend_roles from set to list
• Updated required_permissions and recommend_roles lists to be dynamic based on user tooling selection
• Changed google_project_iam_member block to use a local list for setting iam roles
• Moved csr creation to only occur if use_ci=True
• Updated github_actions config to use source_repo_branch parameter into of default automlops branch

Fixed

• Fixed run local issue regarding dockerfile pathing
• Fixed cloudbuild trigger terraform bug with ignored_files

[1.2.5] - 2023-10-05

Added

Changed

Fixed

• Fixed bug with generating .github/workflows directories

[1.2.4] - 2023-10-03

Added

Changed

• Only will create .github/ directory if using github-actions

Fixed

• Pinned the python docker container step for the cloudbuild yaml to python:3.10, this addresses some dependency issues with the latest python docker image

[1.2.3] - 2023-09-29

Added

• Support for Github Actions integration
• Added workload identity federation integration

Changed

• Updated documentation
• .git and relevant files are now placed inside the AutoMLOps/ base directory

Fixed

• Set default cloud functions registry to the artifact-registry that is created as part of the provision operation

[1.2.2] - 2023-09-25

Added

Changed

• Commented out change from 1.2.1 to set cloud functions default registry to artifact-registry; there appears to be an auth bug with this setting
• Updated documentation

Fixed

• Removed legacy ml.googleapis.com API that is deprecated as of 9/25/2023

[1.2.1] - 2023-09-20

Added

• Support for Vertex AI experiments
• Added in an include statement for the build trigger (only runs on changes to AutoMLOps/**)

Changed

• Set default cloud functions registry to artifact-registry
• Changed recommended roles and permissions for deploying with use_ci=False.

Fixed

• Fixed issues with use_ci=False provision directory not being created properly
• Fixed issues with use_ci=False terraform config

[1.2.0] - 2023-09-19

Added

Major version updates:

• Code is now broken down into 5 main operations: generate, provision, deprovision, deploy, and launchAll
• Uses jinja2 templates for storing code templates and writing to files
• Additional package dependencies using Jinja2 templates and deploying with precheck function
• Provides additional warnings and checks for giving IAM permissions visibility to the user
• Creates a .gitignore by default
• Support for cloud-functions in addition to cloud-run for the submission service
• Added 2 new generated folders: provision/ and services/
• Added naming_prefix parameter to allow for differentiating between AutoMLOps pipelines in the same project
• Significant generalization in terms of tooling (now allows for specifying provisioning_frameworks, deployment_frameworks, etc.)
• Renamed backend src code folder to google_cloud_automlops to avoid naming confusion
• Added enum and config files, which is different than previous approach of class inheritance

Changed

• Updated README.md and documentation
• Templatized code has now been pulled out and placed into jinja2 templates.
• Gitops workflow placed into separate folder and file, will only version AutoMLOps/ directory instead of the whole cwd.
• Reworked deployment workflow and build configuration (submission service and cloud scheduler are now created as part of the provision step).
• Update notebook examples
• Changed wording and branding descriptions
• Significant updates to unit tests

Fixed

• Bugs related to provisioning with terraform

[1.1.4] - 2023-07-25

Added

• Writes .gitkeep to scripts/pipeline_spec directory by default
• Generates a readme.md into generated AutoMLOps codebase now by default

Changed

• Two newlines after functions (linting)
• Parameter mapping (list -> JsonArray, map -> JsonObject)
• Updated documentation: added examples section into main readme, changed package version deps in examples notebooks

Fixed

• Migration issues with pyyaml 5.4.1 since release of cython>3.0; Fixed by updating pyyaml version to 6.0.1.

[1.1.3] - 2023-07-07

Added

• Added a BQML retail notebook example

Changed

• Updated the git workflow to check that the remote is pointing to the correct project id.

Fixed

• Pinned all kfp versions to kfp<2.0.0. to address the recent migration to kfp2+.

Added

• Added in optional parameter for specifying a base_image.
• Created an example notebook that walks the user through a transfer learning example using a GPU.
• Added in a clear_cache function which deletes all files within the tmpfiles directory.

Changed

• Updated readme and implementation guide.
• Long-term change: .tmpfiles/ subdirectory to .AutoMLOps-cache/
• Long-term change: removed use_kfp_spec from parameter lists and switched to determining this at run-time.

Fixed

• Removed redundant code for cloudbuild config generation.
• Updated constants.py file to remove constants no longer being used.
• Fixed custom imports mismatch between kfp spec and custom automlops spec.
• Verified custom_training_jobs_specs works as intended.

Added

• Refactored backend modules to be structured based on frameworks (e.g. kfp, tfx, etc.) and deployments (e.g. cloudbuild, github actions, etc.)
• Added some unit tests for the utils.py module.

Changed

• Moved unit tests to /tests directory.

Added

• New interface for defining AutoMLOps components and Pipelines. Removed the need to call Jupyter cell decorators and replaced them with Python function based decorators instead.
• New feature updates allow for running AutoMLOps outside of Jupyter Notebooks.
• Examples for running AutoMLOps outside of a notebook, as well as an example inferencing pipeline.
• Faster build jobs

Changed

• Updated readme and implementation guide.
• Better logging
• Better handling of requirements.txt (remove dups, infer from pipreqs, option to input explicit versions, sorted order)
• Better versioning of models

Fixed

• Bug that can change current working directory on failure with run_local=True.

Official release on PyPI.

Changed

• Cleaned up wheel and egg files from repo.
• Removed dist/ and build/ directories.

Added feature to allow for accelerated/distributed model training.

Added

• Support for custom training job specs (specifies which resources to use for pipeline jobs).

Changed

• Updated readme and implementation guide.
• New custom_training_job_specs parameter.
• Changed workflow for PipelineBuilder

Fixed

• Bug related to grep substring match for create_resources script.

Reworked process to submit jobs to cloud runner service.

Added

• Cloud Tasks Queue; API enabling, queue creation, and task generation.

Changed

• run() workflow reduced; submission to cloud runner service now takes place as part of the cloudbuild script.
• Creation of schedule job is now part of the cloudbuild script.
• Removed submit_to_runner_svc.sh and create_scheduler.sh.
• Added support for vpc_connectors.

Fixed

• Bug related to waiting for cloudbuild job to complete before submitting to cloud runner service.
• Bug related to elevated IAM privileges in order to authenticate before submitting to cloud runner service.

Official release of AutoMLOps!

AutoMLOps is a service that generates a production ready MLOps pipeline from Jupyter Notebooks, bridging the gap between Data Science and DevOps and accelerating the adoption and use of Vertex AI. The service generates an MLOps codebase for users to customize, and provides a way to build and manage a CI/CD integrated MLOps pipeline from the notebook. AutoMLOps automatically builds a source repo for versioning, cloudbuild configs and triggers, an artifact registry for storing custom components, gs buckets, service accounts and updated IAM privs for running pipelines, enables APIs (cloud Run, Cloud Build, Artifact Registry, etc.), creates a runner service API in Cloud Run for submitting PipelineJobs to Vertex AI, and a Cloud Scheduler job for submitting PipelineJobs on a recurring basis. These automatic integrations empower data scientists to take their experiments to production more quickly, allowing them to focus on what they do best: providing actionable insights through data.