autotab 0.11

optimization of ML pipeline using hierarchical optimization method

autotab

optimize pipeline for any machine learning mdoel using hierarchical optimization method for tabular datasets.

Installation

This package can be installed using pip from pypi using following command

pip install autotab

or using github link for the latest code

python -m pip install git+https://github.com/Sara-Iftikhar/autotab.git

or using setup file, go to folder where this repoitory is downloaded

python setup.py install

Example

Click here to or cick here to

from ai4water.datasets import busan_beach
from skopt.plots import plot_objective
from autotab import OptimizePipeline

data = busan_beach()
input_features = data.columns.tolist()[0:-1]
output_features = data.columns.tolist()[-1:]

transformations = ['minmax', 'zscore', 'log', 'log10', 'sqrt', 'robust', 'quantile', 'none', 'scale']

pl = OptimizePipeline(
    inputs_to_transform=data.columns.tolist()[0:-1],
    parent_iterations=400,
    child_iterations=20,
    parent_algorithm='bayes',
    child_algorithm="random",
    cv_parent_hpo=True,
    eval_metric='mse',
    monitor=['r2', 'nse'],
    input_transformations = transformations,
    output_transformations = transformations,
    models=[ "LinearRegression",
            "LassoLars",
            "Lasso",
            "RandomForestRegressor",
            "HistGradientBoostingRegressor",
             "CatBoostRegressor",
             "XGBRegressor",
             "LGBMRegressor",
             "GradientBoostingRegressor",
             "ExtraTreeRegressor",
             "ExtraTreesRegressor"
             ],

    input_features=data.columns.tolist()[0:-1],
    output_features=data.columns.tolist()[-1:],
    cross_validator={"KFold": {"n_splits": 5}},
    split_random=True,
)

get version information

pl._version_info()

perform optimization

results = pl.fit(data=data, process_results=False)

print optimization report

print(pl.report())

show convergence plot

pl.optimizer_._plot_convergence(save=False)

pl.optimizer_._plot_parallel_coords(figsize=(16, 8), save=False)

_ = pl.optimizer_._plot_distributions(save=False)

pl.optimizer_.plot_importance(save=False)

pl.optimizer_.plot_importance(save=False, plot_type="bar")

_ = plot_objective(results)

pl.optimizer._plot_evaluations(save=False)

pl.optimizer._plot_edf(save=False)

pl.dumbbell_plot(data=data)

pl.dumbbell_plot(data=data, metric_name='r2')

pl.taylor_plot(data=data, save=False, figsize=(6,6))

pl.compare_models()

pl.compare_models(plot_type="bar_chart")

pl.compare_models("r2", plot_type="bar_chart")

get best pipeline with respect to evaluation metric

pl.get_best_pipeline_by_metric('r2')

build fit and evaluate the best pipeline

model = pl.bfe_best_model_from_scratch(data=data)

pl.evaluate_model(model, data=data)

pl.evaluate_model(model, data=data, metric_name='nse')

pl.evaluate_model(model, data=data, metric_name='r2')

get best pipeline with respect to $R^2$

pl.get_best_pipeline_by_metric('r2')

model = pl.bfe_best_model_from_scratch(data=data, metric_name='r2')

pl.evaluate_model(model, data=data, metric_name='r2')

print(f"all results are save in {pl.path} folder")