Pipeline-Bench

Exploring the Search Space

To fully explore the pipeline parameterization, use the following Python code:

from pipeline_bench.lib.core.search_space import get_search_space
from pprint import pprint

# Retrieve the search space
search_space = get_search_space()

# Print the search space
pprint(search_space)

# Print the names of the hyperparameters
hps = search_space.get_hyperparameter_names()
pprint(hps)

The search space consists of several key aspects, some of which are highlighted below:

Classifiers

• adaboost
• bernoulli_nb
• decision_tree
• extra_trees
• gaussian_nb
• gradient_boosting
• k_nearest_neighbors
• lda
• liblinear_svc
• libsvm_svc
• mlp
• multinomial_nb
• passive_aggressive
• qda
• random_forest
• sgd

Feature Preprocessors

• extra_trees_preproc_for_classification
• fast_ica
• feature_agglomeration
• kernel_pca
• kitchen_sinks
• liblinear_svc_preprocessor
• no_preprocessing
• nystroem_sampler
• pca
• polynomial
• random_trees_embedding
• select_percentile_classification
• select_rates_classification

Categorical Encoding

• encoding
• no_encoding
• one_hot_encoding

Rescaling

• minmax
• none
• normalize
• power_transformer
• quantile_transformer
• robust_scaler
• standardize

Contributing to Pipeline-Bench

If you'd like to contribute to Pipeline-Bench, follow the guidelines below:

Managing git submodules

For working with submodules, refer to the git-scm documentation. You can pull changes for Pipeline-Bench and all its submodules (auto-sklearn) using the command:

git pull --recurse-submodules

Optional: Install miniconda and create an environment

wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O install_miniconda.sh
bash install_miniconda.sh -b -p $HOME/.conda  # Change to place of preference
rm install_miniconda.sh

Consider running ~/.conda/bin/conda init or ~/.conda/bin/conda init zsh.

Create the environment and activate it

conda create -n Pipeline-Bench python=3.9
conda activate Pipeline-Bench

Install poetry

First, install poetry, e.g., via

curl -sSL https://install.python-poetry.org | python3 -

Consider appending export PATH="$HOME/.local/bin:$PATH" into ~/.zshrc / ~/.bashrc.

Let poetry take care of all dependencies

poetry install

In case you do do not wish to create any data (use "live" API), run

poetry install --extras "without_data_creation"

To install a new dependency use poetry add dependency and commit the updated pyproject.toml to git.

Activate pre-commit

pre-commit install

Consider appending --no-verify to your urgent commits to disable checks.

Working with git submodules

See the git-scm documentation. In short:

To pull in changes for Pipeline-Bench and all submodules (auto-sklearn) run

git pull --recurse-submodules