2.6.8

ADS

• Fixed a bug in ads.dataset.helper to support Python 3.8 and Python 3.9.

2.6.7

ADS

• Fixed a bug in PyTorchModel. The score.py failed when torch.Tensor was used as input data.
• Added support for flexible shapes for Data Flow Jobs.
• Loading a model from Model Catalog (GenericModel.from_model_catalog()) and Model Deployment (GenericModel.from_model_deployment()) no longer requires a model file name.
• Switched from using cx_Oracle interface to the oracledb driver to connect to Oracle Databases.
• Added support for image attribute for the PyTorchModel.predict() and TensorFlowModel.predict() methods. Images can now be directly passed to the model Deployment predict.

The following APIs are deprecated:

• OracleAutoMLProvider

2.6.6

ADS

• Added SparkPipelineModel model serialization class for fast and easy model deployment.
• Added support for flexible shapes for Jobs and Model Deployments.
• Added support for freeform_tags and defined_tags for Model Deployments.
• Added the populate_schema() method to the GenericModel class. Populate input and output schemas for model artifacts.
• The ADSString was added to the Feature types system. Use the enhanced string class functionalities such as regular expression (RegEx) matching and natural language parsing within Pandas dataframes and series.
• Saving model does not require iPython dependencies

Following APIs are deprecated:

• DatasetFactory.open
• ADSModel.prepare
• ads.common.model_export_util.prepare_generic_model

2.6.5

ADS

• OCI SDK updated from version 2.59.0 to version 2.82.0.

2.6.4

ADS

• Added support for large models with artifact size between 2 and 6 GB. The large models can be saved to the Model Catalog, downloaded from the Model Catalog, and deployed as a Model Deployment resource.
• Added delete() method to the GenericModel class. Deletes models and associated model deployments.
• The Model Input Schema is improved to return features sorted by the order attribute.
• Added user-friendly default names for created Jobs, Model Deployments, and Models.

2.6.3

ADS

• Deprecated the ads.dataflow.DataFlow class. It has been superseded by the ads.jobs.DataFlow class.
• Added prepare_save_deploy() method to the GenericModel class. Prepare model artifacts and deploy the model with one command.
• Added support for binary payloads in model deployment.
• Updated AutoMLModel, GenericModel, LightgbmModel, PyTorchModel, SklearnModel, TensorflowModel, and XgboostModel classes to support binary payloads in model deployment.
• The maximum runtime for a Job can be limited with the with_maximum_runtime_in_minutes() method in the CondaRuntime, DataFlowNotebookRuntime, DataFlowRuntime, GitPythonRuntime, NotebookRuntime, and ScriptRuntime classes.
• The ads.jobs.DataFlow class supports Published conda environments.

2.6.2

ADS

• Added from_model_deployment() method to the GenericModel class. Now you can load a model directly from an existing model deployment.

• Moved dependencies from being default into optional installation groups:

  • all-optional
  • bds
  • boosted
  • data
  • geo
  • notebook
  • onnx
  • opctl
  • optuna
  • tensorflow
  • text
  • torch
  • viz

  Use python3 -m pip install oracle-ads[XXX] where XXX are the group names.

2.6.1

ADS

• Added support for running a container as jobs using ads.jobs.ContainerRuntime.
• The ModelArtifact class is deprecated. Use the model serialization classes (GenericModel, PyTorchModel, SklearnModel, etc.).

2.5.10

ADS

• Added BDSSecretKeeper to store and save configuration parameters to connect to Big Data service to the vault.
• Added the krbcontext and refresh_ticket functions to configure Kerberos authentication for the Big Data service.
• Added authentication options to logging APIs to allow you to pass in the OCI API key configuration or signer.
• Added the configuration file path option to the set_auth method to allow to change the path of the OCI configuration.
• Fixed a bug in AutoML for Ttext datasets.
• Fixed bug in import ads.jobs to notify users installing ADS optional dependencies.
• Fixed a bug in the generated score.py file, where Pandas dataframe's dtypes changed when deserializing. Now you can recover it from the input schema.
• Updated requirements to oci>=2.59.0.

2.5.9

ADS

• Added framework specific model serialization to add more inputs to the generated score.py file.

• Added the following framework-specific model classes:

  • AutoMLModel
  • SKlearnModel
  • XGBoostModel
  • LightGBMModel
  • PyTorchModel
  • TensorFlowModel

• For any framework not included in the preceding list, added another class:

  • GenericModel

• These model classes include methods specific to the frameworks that improve deployment speed. Some example methods are:

  • Prepare (the artifacts)
  • Save (metadata and model to model catalog)
  • Deploy (the models quickly with this method)
  • Predict (perform inference operations)

• Added support to create jobs with managed egress.

• Shortened the time for streaming large number of logs for job run logging.