104 lines
4.6 KiB
Python
104 lines
4.6 KiB
Python
# scipy-yli: Helpful SciPy utilities and recipes
|
|
# Copyright © 2022 Lee Yingtong Li (RunasSudo)
|
|
#
|
|
# This program is free software: you can redistribute it and/or modify
|
|
# it under the terms of the GNU Affero General Public License as published by
|
|
# the Free Software Foundation, either version 3 of the License, or
|
|
# (at your option) any later version.
|
|
#
|
|
# This program is distributed in the hope that it will be useful,
|
|
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
# GNU Affero General Public License for more details.
|
|
#
|
|
# You should have received a copy of the GNU Affero General Public License
|
|
# along with this program. If not, see <https://www.gnu.org/licenses/>.
|
|
|
|
import numpy as np
|
|
import pandas as pd
|
|
import patsy
|
|
from tqdm import tqdm
|
|
|
|
from datetime import datetime
|
|
|
|
from .regress import CategoricalTerm, RegressionResult, SingleTerm, regress
|
|
from .utils import Estimate, check_nan, cols_for_formula, convert_pandas_nullable, fmt_p, formula_factor_ref_category, parse_patsy_term
|
|
|
|
def bootstrap_regress(
|
|
model_class, df, dep, formula, *,
|
|
nan_policy='warn',
|
|
samples=1000,
|
|
**kwargs
|
|
):
|
|
"""
|
|
Fit a statsmodels regression model, using bootstrapping to compute confidence intervals and *p* values
|
|
|
|
:param model_class: See :func:`yli.regress`
|
|
:param df: See :func:`yli.regress`
|
|
:param dep: See :func:`yli.regress`
|
|
:param formula: See :func:`yli.regress`
|
|
:param nan_policy: See :func:`yli.regress`
|
|
:param samples: Number of bootstrap samples to draw
|
|
:type samples: int
|
|
:param kwargs: See :func:`yli.regress`
|
|
|
|
:rtype: :class:`yli.regress.RegressionResult`
|
|
"""
|
|
|
|
# Check for/clean NaNs
|
|
# Following this, we pass nan_policy='raise' to assert no NaNs remaining
|
|
df = df[[dep] + cols_for_formula(formula, df)]
|
|
df = check_nan(df, nan_policy)
|
|
|
|
# Ensure numeric type for dependent variable
|
|
if df[dep].dtype != 'float64':
|
|
df[dep] = df[dep].astype('float64')
|
|
|
|
# Convert pandas nullable types for independent variables as this breaks statsmodels
|
|
df = convert_pandas_nullable(df)
|
|
|
|
# Precompute design matrices
|
|
dmatrices = patsy.dmatrices(dep + ' ~ ' + formula, df, return_type='dataframe')
|
|
|
|
# Fit full model
|
|
full_model = regress(model_class, df, dep, formula, nan_policy='raise', _dmatrices=dmatrices, **kwargs)
|
|
|
|
# Cache reference categories
|
|
ref_categories = {term_name: term.ref_category for term_name, term in full_model.terms.items() if isinstance(term, CategoricalTerm)}
|
|
|
|
# Draw bootstrap samples and regress
|
|
bootstrap_results = []
|
|
for i in tqdm(range(samples)):
|
|
#bootstrap_sample = df.sample(len(df), replace=True)
|
|
#bootstrap_results.append(regress(model_class, bootstrap_sample, dep, formula, nan_policy='raise', _dmatrices=dmatrices, _ref_categories=ref_categories, **kwargs))
|
|
bootstrap_rows = pd.Series(dmatrices[0].index).sample(len(df), replace=True)
|
|
bootstrap_dmatrices = (dmatrices[0].loc[bootstrap_rows], dmatrices[1].loc[bootstrap_rows])
|
|
bootstrap_results.append(regress(model_class, df, dep, formula, nan_policy='raise', _dmatrices=bootstrap_dmatrices, _ref_categories=ref_categories, **kwargs))
|
|
|
|
# Combine bootstrap results
|
|
terms = {}
|
|
for term_name, term in full_model.terms.items():
|
|
if isinstance(term, SingleTerm):
|
|
bootstrap_betas = [r.terms[term_name].beta.point for r in bootstrap_results]
|
|
bootstrap_pvalue = sum(1 for b in bootstrap_betas if b < 0) / len(bootstrap_betas)
|
|
bootstrap_pvalue = 2 * min(bootstrap_pvalue, 1 - bootstrap_pvalue)
|
|
terms[term_name] = SingleTerm(term.raw_name, Estimate(term.beta.point, np.quantile(bootstrap_betas, 0.025), np.quantile(bootstrap_betas, 0.975)), bootstrap_pvalue)
|
|
else:
|
|
categories = {}
|
|
for sub_term_name, sub_term in term.categories.items():
|
|
bootstrap_betas = [r.terms[term_name].categories[sub_term_name].beta.point for r in bootstrap_results if sub_term_name in r.terms[term_name].categories]
|
|
bootstrap_pvalue = sum(1 for b in bootstrap_betas if b < 0) / len(bootstrap_betas)
|
|
bootstrap_pvalue = 2 * min(bootstrap_pvalue, 1 - bootstrap_pvalue)
|
|
categories[sub_term_name] = SingleTerm(sub_term.raw_name, Estimate(sub_term.beta.point, np.quantile(bootstrap_betas, 0.025), np.quantile(bootstrap_betas, 0.975)), bootstrap_pvalue)
|
|
terms[term_name] = CategoricalTerm(categories, term.ref_category)
|
|
|
|
return RegressionResult(
|
|
None,
|
|
full_model.full_name, full_model.model_name, full_model.fit_method,
|
|
dep, full_model.nobs, full_model.dof_model, datetime.now(), full_model.cov_type,
|
|
terms,
|
|
full_model.llf, full_model.llnull,
|
|
full_model.dof_resid, full_model.rsquared, full_model.f_statistic,
|
|
full_model.exp
|
|
)
|