This manual gives an overview of different aspects of auto-sklearn. For each section, we either references examples or give short explanations (click the title to expand text), e.g.

Code examples

We provide examples on using auto-sklearn for multiple use cases ranging from simple classification to advanced uses such as feature importance, parallel runs and customization. They can be found in the Examples.

Material from talks and presentations

We provide resources for talks, tutorials and presentations on auto-sklearn under auto-sklearn-talks

Auto-sklearn 2.0

Auto-sklearn 2.0 includes latest research on automatically configuring the AutoML system itself and contains a multitude of improvements which speed up the fitting the AutoML system. Concretely, Auto-sklearn 2.0 automatically sets the Model selection, decides whether it can use the efficient bandit strategy Successive Halving and uses meta-feature free Portfolios for efficient meta-learning.

auto-sklearn 2.0 has the same interface as regular auto-sklearn and you can use it via

from autosklearn.experimental.askl2 import AutoSklearn2Classifier

A paper describing our advances is available on arXiv.

Resource limits

A crucial feature of auto-sklearn is limiting the resources (memory and time) which the scikit-learn algorithms are allowed to use. Especially for large datasets, on which algorithms can take several hours and make the machine swap, it is important to stop the evaluations after some time in order to make progress in a reasonable amount of time. Setting the resource limits is therefore a tradeoff between optimization time and the number of models that can be tested.

Time and memory limits

While auto-sklearn alleviates manual hyperparameter tuning, the user still has to set memory and time limits. For most datasets a memory limit of 3GB or 6GB as found on most modern computers is sufficient. For the time limits it is harder to give clear guidelines. If possible, a good default is a total time limit of one day, and a time limit of 30 minutes for a single run.

Further guidelines can be found in auto-sklearn/issues/142.

CPU cores

By default, auto-sklearn uses one core. See also Parallel computation on how to configure this.

Managing data compression

Auto-sklearn will attempt to fit the dataset into 1/10th of the memory_limit. This won’t happen unless your dataset is quite large or you have small a memory_limit. This is done using two methods, reducing precision and to subsample. One reason you may want to control this is if you require high precision or you rely on predefined splits for which subsampling does not account for.

To turn off data preprocessing:

    dataset_compression = False

You can specify which of the methods are performed using:

    dataset_compression = { "methods": ["precision", "subsample"] },

You can change the memory allocation for the dataset to a percentage of memory_limit or an absolute amount using:

    dataset_compression = { "memory_allocation": 0.2 },

The default arguments are used when dataset_compression = True are:

    "memory_allocation": 0.1,
    "methods": ["precision", "subsample"]

The full description is given at AutoSklearnClassifier(dataset_compression=...).

The search space

Auto-sklearn by default searches a large space to find a well performing configuration. However, it is also possible to restrict the searchspace:

Restricting the searchspace

The following shows an example of how to exclude all preprocessing methods and restrict the configuration space to only random forests.

import autosklearn.classification
automl = autosklearn.classification.AutoSklearnClassifier(
    include = {
        'classifier': ["random_forest"],
        'feature_preprocessor': ["no_preprocessing"]
), y_train)
predictions = automl.predict(X_test)

Note: The strings used to identify estimators and preprocessors are the filenames without .py.

For a full list please have a look at the source code (in autosklearn/pipeline/components/):

We do also provide an example on how to restrict the classifiers to search over Interpretable models.

Turn off data preprocessing

Data preprocessing includes One-Hot encoding of categorical features, imputation of missing values and the normalization of features or samples. These ensure that the data the gets to the sklearn models is well formed and can be used for training models.

While this is necessary in general, if you’d like to disable this step, please refer to this example.

Turn off feature preprocessing

Feature preprocessing is a single transformer which implements for example feature selection or transformation of features into a different space (i.e. PCA).

This can be turned off by setting include={'feature_preprocessor'=["no_preprocessing"]} as shown in the example above.

Model selection

Auto-sklearn implements different strategies to identify the best performing model. For some use cases it might be necessary to adapt the resampling strategy or define a custom metric:

Use different resampling strategies

Examples for using holdout and cross-validation can be found in example

Use a custom metric

Examples for using a custom metric can be found in example


To get the best performance out of the evaluated models, auto-sklearn uses ensemble selection by Caruana et al. (2004) to build an ensemble based on the models’ prediction for the validation set.

Configure the ensemble building process

The following hyperparameters control how the ensemble is constructed:

  • ensemble_size determines the maximal size of the ensemble. If it is set to zero, no ensemble will be constructed.

  • ensemble_nbest allows the user to directly specify the number of models considered for the ensemble. This hyperparameter can be an integer n, such that only the best n models are used in the final ensemble. If a float between 0.0 and 1.0 is provided, ensemble_nbest would be interpreted as a fraction suggesting the percentage of models to use in the ensemble building process (namely, if ensemble_nbest is a float, library pruning is implemented as described in Caruana et al. (2006)).

  • max_models_on_disc defines the maximum number of models that are kept on the disc, as a mechanism to control the amount of disc space consumed by auto-sklearn. Throughout the automl process, different individual models are optimized, and their predictions (and other metadata) is stored on disc. The user can set the upper bound on how many models are acceptable to keep on disc, yet this variable takes priority in the definition of the number of models used by the ensemble builder (that is, the minimum of ensemble_size, ensemble_nbest and max_models_on_disc determines the maximal amount of models used in the ensemble). If set to None, this feature is disabled.

Inspect the final ensemble

The results obtained from the final ensemble can be printed by calling show_models(). The auto-sklearn ensemble is composed of scikit-learn models that can be inspected as exemplified in the Example Obtain run information.

Fit ensemble post-hoc

To use a single core only, it is possible to build ensembles post-hoc. An example on how to do this (first searching for individual models, and then building an ensemble from them) can be seen in Sequential Usage.

Inspecting the results

auto-sklearn allows users to inspect the training results and statistics. Assume we have a fitted estimator:

import autosklearn.classification
automl = autosklearn.classification.AutoSklearnClassifier(), y_train)

auto-sklearn offers the following ways to inspect the results

Basic statistics

sprint_statistics() is a method that prints the name of the dataset, the metric used, and the best validation score obtained by running auto-sklearn. It additionally prints the number of both successful and unsuccessful algorithm runs.

Performance over Time

performance_over_time_ returns a DataFrame containing the models performance over time data, which can be used for plotting directly (Here is an example: Performance-over-time plot).

        title='Auto-sklearn accuracy over time',
Evaluated models

The results obtained from the final ensemble can be printed by calling show_models().


automl.leaderboard() shows the ensemble members, check the docs for using leaderboard for getting information on all runs.


cv_results_ returns a dict with keys as column headers and values as columns, that can be imported into a pandas DataFrame.

Parallel computation

In it’s default mode, auto-sklearn uses one core and interleaves ensemble building with evaluating new configurations.

Parallelization with Dask

Nevertheless, auto-sklearn also supports parallel Bayesian optimization via the use of Dask.distributed. By providing the arguments n_jobs to the estimator construction, one can control the number of cores available to auto-sklearn (As shown in the Example Parallel Usage on a single machine). Distributed processes are also supported by providing a custom client object to auto-sklearn like in the Example: Parallel Usage: Spawning workers from the command line. When multiple cores are available, auto-sklearn will create a worker per core, and use the available workers to both search for better machine learning models as well as building an ensemble with them until the time resource is exhausted.

Note: auto-sklearn requires all workers to have access to a shared file system for storing training data and models.

auto-sklearn employs threadpoolctl to control the number of threads employed by scientific libraries like numpy or scikit-learn. This is done exclusively during the building procedure of models, not during inference. In particular, auto-sklearn allows each pipeline to use at most 1 thread during training. At predicting and scoring time this limitation is not enforced by auto-sklearn. You can control the number of resources employed by the pipelines by setting the following variables in your environment, prior to running auto-sklearn:

$ export MKL_NUM_THREADS=1
$ export OMP_NUM_THREADS=1

For further information about how scikit-learn handles multiprocessing, please check the Parallelism, resource management, and configuration documentation from the library.


Supported input types

auto-sklearn can accept targets for the following tasks (more details on Sklearn algorithms):

  • Binary Classification

  • Multiclass Classification

  • Multilabel Classification

  • Regression

  • Multioutput Regression

You can provide feature and target training pairs (X_train/y_train) to auto-sklearn to fit an ensemble of pipelines as described in the next section. This X_train/y_train dataset must belong to one of the supported formats: np.ndarray, pd.DataFrame, scipy.sparse.csr_matrix and python lists. Optionally, you can measure the ability of this fitted model to generalize to unseen data by providing an optional testing pair (X_test/Y_test). For further details, please refer to the Example Performance-over-time plot.

Regarding the features, there are multiple things to consider:

  • Providing a X_train/X_test numpy array with the optional flag feat_type. For further details, you can check the Example Feature Types.

  • You can provide a pandas DataFrame with properly formatted columns. If a column has numerical dtype, auto-sklearn will not encode it and it will be passed directly to scikit-learn. auto-sklearn supports both categorical or string as column type. Please ensure that you are using the correct dtype for your task. By default auto-sklearn treats object and string columns as strings and encodes the data using sklearn.feature_extraction.text.CountVectorizer

  • If your data contains categorical values (in the features or targets), ensure that you explicitly label them as categorical. Data labeled as categorical is encoded by using a sklearn.preprocessing.LabelEncoder for unidimensional data and a sklearn.preprodcessing.OrdinalEncoder for multidimensional data.

  • For further details on how to properly encode your data, you can check the Pandas Example Working with categorical data). If you are working with time series, it is recommended that you follow this approach Working with time data.

  • If you prefer not using the string option at all you can disable this option. In this case objects, strings and categorical columns are encoded as categorical.

import autosklearn.classification
automl = autosklearn.classification.AutoSklearnClassifier(allow_string_features=False), y_train)

Regarding the targets (y_train/y_test), if the task involves a classification problem, such features will be automatically encoded. It is recommended to provide both y_train and y_test during fit, so that a common encoding is created between these splits (if only y_train is provided during fit, the categorical encoder will not be able to handle new classes that are exclusive to y_test). If the task is regression, no encoding happens on the targets.

Model persistence

auto-sklearn is mostly a wrapper around scikit-learn. Therefore, it is possible to follow the persistence Example from scikit-learn.

Vanilla auto-sklearn

In order to obtain vanilla auto-sklearn as used in Efficient and Robust Automated Machine Learning set ensemble_size=1, initial_configurations_via_metalearning=0 and allow_string_features=False:

import autosklearn.classification
automl = autosklearn.classification.AutoSklearnClassifier(

An ensemble of size one will result in always choosing the current best model according to its performance on the validation set. Setting the initial configurations found by meta-learning to zero makes auto-sklearn use the regular SMAC algorithm for suggesting new hyperparameter configurations.

Early stopping and Callbacks

By using the parameter get_trials_callback, we can get access to the results of runs as they occur. See this example Early Stopping And Callbacks for more!