ClassificationJob Class

The local path or file stream to write the YAML content to. If dest is a file path, a new file will be created. If dest is an open file, the file will be written to directly.

Training data.

Column name of the target column.

Weight column name, defaults to None

Validation data, defaults to None

Validation data size, defaults to None

n_cross_validations, defaults to None

cv_split_column_names, defaults to None

Test data, defaults to None

Test data size, defaults to None

A list of transformer names to be blocked during featurization, defaults to None

A dictionary of column names and feature types used to update column purpose , defaults to None

Three character ISO 639-3 code for the language(s) contained in the dataset. Languages other than English are only supported if you use GPU-enabled compute. The language_code 'mul' should be used if the dataset contains multiple languages. To find ISO 639-3 codes for different languages, please refer to https://en.wikipedia.org/wiki/List_of_ISO_639-3_codes, defaults to None

A dictionary of transformer and corresponding customization parameters , defaults to None

"off", "auto", defaults to "auto", defaults to None

Whether to include DNN based feature engineering methods, defaults to None

Whether to enable early termination if the score is not improving in the short term, defaults to None.

Early stopping logic:

• No early stopping for first 20 iterations (landmarks).

• Early stopping window starts on the 21st iteration and looks for early_stopping_n_iters iterations

  (currently set to 10). This means that the first iteration where stopping can occur is the 31st.

• AutoML still schedules 2 ensemble iterations AFTER early stopping, which might result in higher scores.

• Early stopping is triggered if the absolute value of best score calculated is the same for past

  early_stopping_n_iters iterations, that is, if there is no improvement in score for early_stopping_n_iters iterations.

Target score for experiment. The experiment terminates after this score is reached. If not specified (no criteria), the experiment runs until no further progress is made on the primary metric. For for more information on exit criteria, see this article , defaults to None

This is the maximum number of iterations that would be executed in parallel. The default value is 1.

• AmlCompute clusters support one iteration running per node. For multiple AutoML experiment parent runs

  executed in parallel on a single AmlCompute cluster, the sum of the max_concurrent_trials values for all experiments should be less than or equal to the maximum number of nodes. Otherwise, runs will be queued until nodes are available.

• DSVM supports multiple iterations per node. max_concurrent_trials should

  be less than or equal to the number of cores on the DSVM. For multiple experiments run in parallel on a single DSVM, the sum of the max_concurrent_trials values for all experiments should be less than or equal to the maximum number of nodes.

• Databricks - max_concurrent_trials should be less than or equal to the number of

  worker nodes on Databricks.

max_concurrent_trials does not apply to local runs. Formerly, this parameter was named concurrent_iterations.

The maximum number of threads to use for a given training iteration. Acceptable values:

• Greater than 1 and less than or equal to the maximum number of cores on the compute target.

• Equal to -1, which means to use all the possible cores per iteration per child-run.

• Equal to 1, the default.

[Experimental] The maximum number of nodes to use for distributed training.

• For forecasting, each model is trained using max(2, int(max_nodes / max_concurrent_trials)) nodes.

• For classification/regression, each model is trained using max_nodes nodes.

Note- This parameter is in public preview and might change in future.

The total number of different algorithm and parameter combinations to test during an automated ML experiment. If not specified, the default is 1000 iterations.

Maximum amount of time in minutes that all iterations combined can take before the experiment terminates. If not specified, the default experiment timeout is 6 days. To specify a timeout less than or equal to 1 hour, make sure your dataset's size is not greater than 10,000,000 (rows times column) or an error results, defaults to None

Maximum time in minutes that each iteration can run for before it terminates. If not specified, a value of 1 month or 43200 minutes is used, defaults to None

Whether to enable or disable enforcing the ONNX-compatible models. The default is False. For more information about Open Neural Network Exchange (ONNX) and Azure Machine Learning,see this article.

Whether to include DNN based models during model selection. However, the default is True for DNN NLP tasks, and it's False for all other AutoML tasks.

Whether to enable explaining the best AutoML model at the end of all AutoML training iterations. For more information, see Interpretability: model explanations in automated machine learning. , defaults to None

Whether to enable/disable StackEnsemble iteration. If enable_onnx_compatible_models flag is being set, then StackEnsemble iteration will be disabled. Similarly, for Timeseries tasks, StackEnsemble iteration will be disabled by default, to avoid risks of overfitting due to small training set used in fitting the meta learner. For more information about ensembles, see Ensemble configuration , defaults to None

Whether to enable/disable VotingEnsemble iteration. For more information about ensembles, see Ensemble configuration , defaults to None

Settings for StackEnsemble iteration, defaults to None

During VotingEnsemble and StackEnsemble model generation, multiple fitted models from the previous child runs are downloaded. Configure this parameter with a higher value than 300 secs, if more time is needed, defaults to None

A list of model names to search for an experiment. If not specified, then all models supported for the task are used minus any specified in blocked_training_algorithms or deprecated TensorFlow models, defaults to None

A list of algorithms to ignore for an experiment, defaults to None

[Experimental] The training mode to use. The possible values are-

• distributed- enables distributed training for supported algorithms.

• non_distributed- disables distributed training.

• auto- Currently, it is same as non_distributed. In future, this might change.

Note: This parameter is in public preview and may change in future.