substrat-automl 0.0.4

A Python package for automated machine learning tasks with genetic algorithm-based dataset summarization.

SubStrat Package

SubStrat is a Python package designed to provide a substrate for automated machine learning tasks. It comes integrated with functionalities from the popular AutoSklearn library, and also includes a genetic algorithm-based dataset summarization utility.

SubStratis based on The SubStart Article

Features

• Automated Machine Learning (AutoML): Using the power of the AutoSklearn library, users can seamlessly train and fine-tune machine learning models on their dataset.

• Genetic Dataset Summarization: SubStrat includes a genetic algorithm-based approach to summarize datasets, providing concise data representations while retaining vital information.

SubStrat Flow

• Run genetic algorithm-based to find sub set dataset that yet represet the full size dataset.
• On the sub set dataset runs full search of Automl.
• Extrat the model with the highet score.
• Run another time the automl to finetune the hyper-parameters fot the specific model.
• Returns the classifier

Very recomended to use venv.

python3 -m venv subsrat_vev

Installing the SubStrat package.

pip install substart-automl

Usage

from SubStrat import SubStrat

# Initialize SubStrat with a dataset and target column
s = SubStrat(dataset=my_dataset, target_col_name='target')
# Excute SubStrat flow
cls = s.run()

see basic example here

Classes

SubStrat

Provides the primary interface for the AutoML functionalities.

Attributes:

• dataset: Input dataset (pandas DataFrame).
• target_col_name: Name of the target column in the dataset.
• input_classifier: Classifier instance (optional). Defaults to an instance from AutoSklearn.
• summary_algorithm: Algorithm to summarize the dataset. Defaults to GeneticSubAlgorithmn.
• desired_accuracy:Desired accuracy for the output classifier.

Methods:

• run(): Executes the SubStrat flow, and returns AutoSklearnClassifier.

GeneticSubAlgorithmn

Implements the genetic algorithm for dataset summarization.

Attributes:

• dataset: The original dataset that needs to be summarized.
• target_column_name: The name of the target column in the dataset.
• sub_row_size: The number of rows for the subset of the dataset (summary). If not provided, it will be calculated based on a predefined rule.
• sub_col_size: The number of columns for the subset of the dataset (summary). If not provided, it will be calculated based on a predefined rule.
• population_size: The number of individuals in the population for the genetic algorithm.
• fitness: The fitness function used in the genetic algorithm. If not provided, a default fitness function will be used.
• selection: The selection operator used in the genetic algorithm. If not provided, a default selection operator will be used.
• mutation_rate: The mutation rate used in the genetic algorithm.
• num_generation: The number of generations the genetic algorithm will run for.
• init_pop: The algorithm used to initialize the population for the genetic algorithm. If not provided, a default algorithm will be used.
• stagnation_limit: Number of generations without improvement in best gene score before stopping.
• time_limit: Maximum time in seconds the run function can execute.

Methods:

• run(): Executes the genetic algorithm and returns the best subset of the dataset.

futute features

• Add verbose mode.
• Add the option to use more AutoML frameworks, like TPOT.
• Make SubStrat more configable by the user.
• Make the UX more friendly.

Citing information

Teddy Lazebnik, Amit Somech, and Abraham Itzhak Weinberg. 2022. SubStrat: A Subset-Based Optimization Strategy for Faster AutoML. Proc. VLDB Endow. 16, 4 (December 2022), 772–780. https://doi.org/10.14778/3574245.3574261