AutoMLConfig class

path

str

The full path to the Azure Machine Learning project folder. If not specified, the default is to use the current directory or ".".

iterations

int

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

primary_metric

str or azureml.train.automl.constants.Metric

The metric that Automated Machine Learning will optimize for model selection. Automated Machine Learning collects more metrics than it can optimize. You can use get_primary_metrics(task) to get a list of valid metrics for your given task. For more information on how metrics are calculated, see https://docs.microsoft.com/azure/machine-learning/how-to-configure-auto-train#primary-metric.

If not specified, accuracy is used for classification tasks, normalized root mean squared is used for forecasting and regression tasks, accuracy is used for image classification and image multi label classification, and mean average precision is used for image object detection.

compute_target

AbstractComputeTarget

The Azure Machine Learning compute target to run the Automated Machine Learning experiment on. See https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-remote for more information on compute targets.

spark_context

SparkContext

The Spark context. Only applicable when used inside Azure Databricks/Spark environment.

X

DataFrame or ndarray or Dataset or DatasetDefinition or TabularDataset

The training features to use when fitting pipelines during an experiment. This setting is being deprecated. Please use training_data and label_column_name instead.

y

DataFrame or ndarray or Dataset or DatasetDefinition or TabularDataset

The training labels to use when fitting pipelines during an experiment. This is the value your model will predict. This setting is being deprecated. Please use training_data and label_column_name instead.

sample_weight

DataFrame or ndarray or TabularDataset

The weight to give to each training sample when running fitting pipelines, each row should correspond to a row in X and y data.

Specify this parameter when specifying X.

X_valid

DataFrame or ndarray or Dataset or DatasetDefinition or TabularDataset

Validation features to use when fitting pipelines during an experiment.

If specified, then y_valid or sample_weight_valid must also be specified.

y_valid

DataFrame or ndarray or Dataset or DatasetDefinition or TabularDataset

Validation labels to use when fitting pipelines during an experiment.

Both X_valid and y_valid must be specified together.

sample_weight_valid

DataFrame or ndarray or TabularDataset

The weight to give to each validation sample when running scoring pipelines, each row should correspond to a row in X and y data.

Specify this parameter when specifying X_valid.

cv_splits_indices

List[List[ndarray]]

Indices where to split training data for cross validation. Each row is a separate cross fold and within each crossfold, provide 2 numpy arrays, the first with the indices for samples to use for training data and the second with the indices to use for validation data. i.e., [[t1, v1], [t2, v2], ...] where t1 is the training indices for the first cross fold and v1 is the validation indices for the first cross fold.

To specify existing data as validation data, use validation_data. To let AutoML extract validation data out of training data instead, specify either n_cross_validations or validation_size. Use cv_split_column_names if you have cross validation column(s) in training_data.

validation_size

float

What fraction of the data to hold out for validation when user validation data is not specified. This should be between 0.0 and 1.0 non-inclusive.

Specify validation_data to provide validation data, otherwise set n_cross_validations or validation_size to extract validation data out of the specified training data. For custom cross validation fold, use cv_split_column_names.

For more information, see Configure data splits and cross-validation in automated machine learning.

n_cross_validations

int

How many cross validations to perform when user validation data is not specified.

Specify validation_data to provide validation data, otherwise set n_cross_validations or validation_size to extract validation data out of the specified training data. For custom cross validation fold, use cv_split_column_names.

For more information, see Configure data splits and cross-validation in automated machine learning.

y_min

float

Minimum value of y for a regression experiment. The combination of y_min and y_max are used to normalize test set metrics based on the input data range. If not specified, the minimum value is inferred from the data.

y_max

float

Maximum value of y for a regression experiment. The combination of y_min and y_max are used to normalize test set metrics based on the input data range. If not specified, the maximum value is inferred from the data.

num_classes

int

The number of classes in the label data for a classification experiment.

featurization

str or FeaturizationConfig

'auto' / 'off' / FeaturizationConfig Indicator for whether featurization step should be done automatically or not, or whether customized featurization should be used. Note: If the input data is sparse, featurization cannot be turned on.

Column type is automatically detected. Based on the detected column type preprocessing/featurization is done as follows:

• Categorical: Target encoding, one hot encoding, drop high cardinality categories, impute missing values.

• Numeric: Impute missing values, cluster distance, weight of evidence.

• DateTime: Several features such as day, seconds, minutes, hours etc.

• Text: Bag of words, pre-trained Word embedding, text target encoding.

More details can be found in the article Configure automated ML experiments in Python.

To customize featurization step, provide a FeaturizationConfig object. Customized featurization currently supports blocking a set of transformers, updating column purpose, editing transformer parameters, and dropping columns. For more information, see Customize feature engineering.

Note: Timeseries features are handled separately when the task type is set to forecasting independent of this parameter.

max_cores_per_iteration

int

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.

max_concurrent_iterations

int

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

• AmlCompute clusters support one interation running per node. For multiple AutoML experiment parent runs executed in parallel on a single AmlCompute cluster, the sum of the max_concurrent_iterations 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_iterations 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_iterations values for all experiments should be less than or equal to the maximum number of nodes.

• Databricks - max_concurrent_iterations should be less than or equal to the number of worker nodes on Databricks.

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

iteration_timeout_minutes

int

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.

mem_in_mb

int

Maximum memory usage that each iteration can run for before it terminates. If not specified, a value of 1 PB or 1073741824 MB is used.

enforce_time_on_windows

bool

Whether to enforce a time limit on model training at each iteration on Windows. The default is True. If running from a Python script file (.py), see the documentation for allowing resource limits on Windows.

experiment_timeout_hours

float

Maximum amount of time in hours that all iterations combined can take before the experiment terminates. Can be a decimal value like 0.25 representing 15 minutes. 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.

experiment_exit_score

float

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 primart metric. For for more information on exit criteria, see this article.

enable_early_stopping

bool

Whether to enable early termination if the score is not improving in the short term. The default is False.

Default behavior for stopping criteria:

• If iteration and experiment timeout are not specified, then early stopping is turned on and

  experiment_timeout = 6 days, num_iterations = 1000.

• If experiment timeout is specified, then early_stopping = off, num_iterations = 1000.

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.

blocked_models

list(str) or list(Classification) for classification task, or list(Regression) for regression task, or list(Forecasting) for forecasting task

A list of algorithms to ignore for an experiment. If enable_tf is False, TensorFlow models are included in blocked_models.

blacklist_models

list(str) or list(Classification) for classification task, or list(Regression) for regression task, or list(Forecasting) for forecasting task

Deprecated parameter, use blocked_models instead.

exclude_nan_labels

bool

Whether to exclude rows with NaN values in the label. The default is True.

verbosity

int

The verbosity level for writing to the log file. The default is INFO or 20. Acceptable values are defined in the Python logging library.

enable_tf

bool

Deprecated parameter to enable/disable Tensorflow algorithms. The default is False.

model_explainability

bool

Whether to enable explaining the best AutoML model at the end of all AutoML training iterations. The default is True. For more information, see Interpretability: model explanations in automated machine learning.

allowed_models

list(str) or list(Classification) for classification task, or list(Regression) for regression task, or list(Forecasting) for forecasting task

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_models or deprecated TensorFlow models. The supported models for each task type are described in the SupportedModels class.

whitelist_models

list(str) or list(Classification) for classification task, or list(Regression) for regression task, or list(Forecasting) for forecasting task

Deprecated parameter, use allowed_models instead.

enable_onnx_compatible_models

bool

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.

time_column_name

str

The name of the time column. This parameter is required when forecasting to specify the datetime column in the input data used for building the time series and inferring its frequency.

max_horizon

int

The desired maximum forecast horizon in units of time-series frequency. The default value is 1.

Units are based on the time interval of your training data, e.g., monthly, weekly that the forecaster should predict out. When task type is forecasting, this parameter is required. For more information on setting forecasting parameters, see Auto-train a time-series forecast model.

grain_column_names

str or list(str)

The names of columns used to group a timeseries. It can be used to create multiple series. If grain is not defined, the data set is assumed to be one time-series. This parameter is used with task type forecasting.

drop_column_names

str or list(str)

The names of columns to drop for forecasting tasks. To customize drop columns for classification and regression tasks, use the featurization parameter.

target_lags

int or list(int)

The number of past periods to lag from the target column. The default is 1.

When forecasting, this parameter represents the number of rows to lag the target values based on the frequency of the data. This is represented as a list or single integer. Lag should be used when the relationship between the independent variables and dependant variable do not match up or correlate by default. For example, when trying to forecast demand for a product, the demand in any month may depend on the price of specific commodities 3 months prior. In this example, you may want to lag the target (demand) negatively by 3 months so that the model is training on the correct relationship. For more information, see Auto-train a time-series forecast model.

feature_lags

str

Flag for generating lags for the numeric features

target_rolling_window_size

int

The number of past periods used to create a rolling window average of the target column.

When forecasting, this parameter represents n historical periods to use to generate forecasted values, <= training set size. If omitted, n is the full training set size. Specify this parameter when you only want to consider a certain amount of history when training the model.

country_or_region

str

The country/region used to generate holiday features. These should be ISO 3166 two-letter country/region code, for example 'US' or 'GB'.

use_stl

str

Configure STL Decomposition of the time-series target column. use_stl can take three values: None (default) - no stl decomposition, 'season' - only generate season component and season_trend - generate both season and trend components.

seasonality

int

Set time series seasonality. If seasonality is set to -1, it will be inferred. If use_stl is not set, this parameter will not be used.

freq

str

The time series data set frequency.

When forecasting this parameter represents the period with which the events are supposed to happen, for example daily, weekly, yearly, etc. The frequency needs to be a pandas offset alias. Please refer to pandas documentation for more information: https://pandas.pydata.org/pandas-docs/stable/user_guide/timeseries.html#dateoffset-objects

enable_voting_ensemble

bool

Whether to enable/disable VotingEnsemble iteration. The default is True. For more information about ensembles, see Ensemble configuration.

enable_stack_ensemble

Optional[bool]

Whether to enable/disable StackEnsemble iteration. The default is None. 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.

debug_log

str

The log file to write debug information to. If not specified, 'automl.log' is used.

training_data

DataFrame or Dataset or DatasetDefinition or TabularDataset

The training data to be used within the experiment. It should contain both training features and a label column (optionally a sample weights column). If training_data is specified, then the label_column_name parameter must also be specified.

training_data was introduced in version 1.0.81.

validation_data

DataFrame or Dataset or DatasetDefinition or TabularDataset

The validation data to be used within the experiment. It should contain both training features and label column (optionally a sample weights column). If validation_data is specified, then training_data and label_column_name parameters must be specified.

validation_data was introduced in version 1.0.81. For more information, see Configure data splits and cross-validation in automated machine learning.

label_column_name

Union[str, int]

The name of the label column. If the input data is from a pandas.DataFrame which doesn't have column names, column indices can be used instead, expressed as integers.

This parameter is applicable to training_data and validation_data parameters. label_column_name was introduced in version 1.0.81.

weight_column_name

Union[str, int]

The name of the sample weight column. Automated ML supports a weighted column as an input, causing rows in the data to be weighted up or down. If the input data is from a pandas.DataFrame which doesn't have column names, column indices can be used instead, expressed as integers.

This parameter is applicable to training_data and validation_data parameters. weight_column_names was introduced in version 1.0.81.

cv_split_column_names

list(str)

List of names of the columns that contain custom cross validation split. Each of the CV split columns represents one CV split where each row are either marked 1 for training or 0 for validation.

This parameter is applicable to training_data parameter for custom cross validation purposes. cv_split_column_names was introduced in version 1.6.0

Use either cv_split_column_names or cv_splits_indices.

For more information, see Configure data splits and cross-validation in automated machine learning.

enable_local_managed

bool

flag whether to allow local managed runs

enable_dnn

bool

Whether to include DNN based models during model selection. The default is False.