azureml_automl_classification_experiment_quickstart_with_iris_dataset.py

1import logging
2from azureml.core import Workspace, Experiment
3from azureml.train.automl import AutoMLConfig
4
5# 1. Connect to Azure ML Workspace
6ws = Workspace.from_config()
7
8# 2. Define the training data (Example: using a TabularDataset)
9# For this example, we assume a dataset 'iris_data' is already registered
10dataset = ws.datasets.get("iris_data")
11training_data = dataset
12
13# 3. Configure the AutoML Experiment
14# Note: azureml-automl-core provides the underlying logic for AutoMLConfig
15automl_settings = {
16    "iteration_timeout_minutes": 10,
17    "experiment_timeout_hours": 0.3,
18    "enable_early_stopping": True,
19    "primary_metric": 'accuracy',
20    "featurization": 'auto',
21    "verbosity": logging.INFO,
22    "n_cross_validations": 5
23}
24
25automl_config = AutoMLConfig(task='classification',
26                             debug_log='automl_errors.log',
27                             training_data=training_data,
28                             label_column_name="species",
29                             **automl_settings)
30
31# 4. Submit the Experiment
32experiment = Experiment(ws, "automl_classification_quickstart")
33local_run = experiment.submit(automl_config, show_output=True)
34
35# 5. Retrieve the best model
36best_run, fitted_model = local_run.get_output()
37print(best_run)
38print(fitted_model)