Logging and debugging

This example shows how to provide a custom logging configuration to auto-sklearn. We will be fitting 2 pipelines and showing any INFO-level msg on console. Even if you do not provide a logging_configuration, autosklearn creates a log file in the temporal working directory. This directory can be specified via the tmp_folder as exemplified below.

This example also highlights additional information about auto-sklearn internal directory structure.

import pathlib

import sklearn.datasets
import sklearn.metrics
import sklearn.model_selection

import autosklearn.classification

Data Loading

Load kr-vs-kp dataset from https://www.openml.org/d/3

X, y = data = sklearn.datasets.fetch_openml(data_id=3, return_X_y=True, as_frame=True)

X_train, X_test, y_train, y_test = \
     sklearn.model_selection.train_test_split(X, y, random_state=1)

Create a logging config

auto-sklearn uses a default logging config We will instead create a custom one as follows:

logging_config = {
    'version': 1,
    'disable_existing_loggers': True,
    'formatters': {
        'custom': {
            # More format options are available in the official
            # `documentation <https://docs.python.org/3/howto/logging-cookbook.html>`_
            'format': '%(asctime)s - %(name)s - %(levelname)s - %(message)s'
        }
    },

    # Any INFO level msg will be printed to the console
    'handlers': {
        'console': {
            'level': 'INFO',
            'formatter': 'custom',
            'class': 'logging.StreamHandler',
            'stream': 'ext://sys.stdout',
        },
    },

    'loggers': {
        '': {  # root logger
            'level': 'DEBUG',
        },
        'Client-EnsembleBuilder': {
            'level': 'DEBUG',
            'handlers': ['console'],
        },
    },
}

Build and fit a classifier

cls = autosklearn.classification.AutoSklearnClassifier(
    time_left_for_this_task=30,
    # Bellow two flags are provided to speed up calculations
    # Not recommended for a real implementation
    initial_configurations_via_metalearning=0,
    smac_scenario_args={'runcount_limit': 2},
    # Pass the config file we created
    logging_config=logging_config,
    # *auto-sklearn* generates temporal files under tmp_folder
    tmp_folder='./tmp_folder',
    # By default tmp_folder is deleted. We will preserve it
    # for debug purposes
    delete_tmp_folder_after_terminate=False,
)
cls.fit(X_train, y_train, X_test, y_test)

# *auto-sklearn* generates intermediate files which can be of interest
# Dask multiprocessing information. Useful on multi-core runs:
#   * tmp_folder/distributed.log
# The individual fitted estimators are written to disk on:
#   * tmp_folder/.auto-sklearn/runs
# SMAC output is stored in this directory.
# For more info, you can check the `SMAC documentation <https://github.com/automl/SMAC3>`_
#   * tmp_folder/smac3-output
# Auto-sklearn always outputs to this log file
# tmp_folder/AutoML*.log
for filename in pathlib.Path('./tmp_folder').glob('*'):
    print(filename)

Out:

2022-04-24 14:27:21,946 - Client-EnsembleBuilder - INFO - DummyFuture: ([{'Timestamp': Timestamp('2022-04-24 14:27:21.933853'), 'ensemble_optimization_score': 0.9886363636363636, 'ensemble_test_score': 0.9899874843554443}], 50, None, None, None)/SingleThreadedClient() Started Ensemble builder job at 2022.04.24-14.27.21 for iteration 0.
2022-04-24 14:27:26,087 - Client-EnsembleBuilder - INFO - DummyFuture: ([{'Timestamp': Timestamp('2022-04-24 14:27:26.073505'), 'ensemble_optimization_score': 0.9911616161616161, 'ensemble_test_score': 0.9899874843554443}], 50, None, None, None)/SingleThreadedClient() Started Ensemble builder job at 2022.04.24-14.27.26 for iteration 1.
tmp_folder/distributed.log
tmp_folder/.auto-sklearn
tmp_folder/space.json
tmp_folder/AutoML(1):a4ac39ac-c3da-11ec-8876-93d4b98a61f7.log
tmp_folder/smac3-output