scipy-yli/yli/sig_tests.py

#   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 pandas as pd
from scipy import stats
import statsmodels.api as sm

import functools
import warnings

from .utils import Estimate, as_2groups, check_nan, fmt_p_html, fmt_p_text

# ----------------
# Student's t test

class TTestResult:
	"""
	Result of a Student's t test
	
	delta: Mean difference
	"""
	
	def __init__(self, statistic, dof, pvalue, delta, delta_direction):
		self.statistic = statistic
		self.dof = dof
		self.pvalue = pvalue
		self.delta = delta
		self.delta_direction = delta_direction
	
	def _repr_html_(self):
		return '<i>t</i>({:.0f}) = {:.2f}; <i>p</i> {}<br><i>δ</i> (95% CI) = {}, {}'.format(self.dof, self.statistic, fmt_p_html(self.pvalue), self.delta.summary(), self.delta_direction)
	
	def summary(self):
		return 't({:.0f}) = {:.2f}; p {}\nδ (95% CI) = {}, {}'.format(self.dof, self.statistic, fmt_p_text(self.pvalue), self.delta.summary(), self.delta_direction)

def ttest_ind(df, dep, ind, *, nan_policy='warn'):
	"""Perform an independent-sample Student's t test"""
	
	# Check for/clean NaNs
	df = check_nan(df[[ind, dep]], nan_policy)
	
	# Ensure 2 groups for ind
	group1, data1, group2, data2 = as_2groups(df, dep, ind)
	
	# Do t test
	# Use statsmodels rather than SciPy because this provides the mean difference automatically
	d1 = sm.stats.DescrStatsW(data1)
	d2 = sm.stats.DescrStatsW(data2)
	
	cm = sm.stats.CompareMeans(d1, d2)
	statistic, pvalue, dof = cm.ttest_ind()
	
	delta = d1.mean - d2.mean
	ci0, ci1 = cm.tconfint_diff()
	
	# t test is symmetric so take absolute values
	return TTestResult(
		statistic=abs(statistic), dof=dof, pvalue=pvalue,
		delta=abs(Estimate(delta, ci0, ci1)),
		delta_direction=('{0} > {1}' if d1.mean > d2.mean else '{1} > {0}').format(group1, group2))

# -----------------
# Mann-Whitney test

class MannWhitneyResult:
	"""
	Result of a Mann-Whitney test
	
	brunnermunzel: BrunnerMunzelResult on same data
	"""
	
	def __init__(self, statistic, pvalue, rank_biserial, direction, brunnermunzel=None):
		self.statistic = statistic
		self.pvalue = pvalue
		self.rank_biserial = rank_biserial
		self.direction = direction
		self.brunnermunzel = brunnermunzel
	
	def _repr_html_(self):
		line1 = '<i>U</i> = {:.1f}; <i>p</i> {}<br><i>r</i> = {:.2f}, {}'.format(self.statistic, fmt_p_html(self.pvalue), self.rank_biserial, self.direction)
		if self.brunnermunzel:
			return line1 + '<br>' + self.brunnermunzel._repr_html_()
		else:
			return line1
	
	def summary(self):
		line1 = 'U = {:.1f}; p {}\nr = {}, {}'.format(self.statistic, fmt_p_text(self.pvalue), self.rank_biserial, self.direction)
		if self.brunnermunzel:
			return line1 + '\n' + self.brunnermunzel.summary()
		else:
			return line1

class BrunnerMunzelResult:
	"""Result of a Brunner-Munzel test"""
	
	def __init__(self, statistic, pvalue):
		self.statistic = statistic
		self.pvalue = pvalue
	
	def _repr_html_(self):
		return '<i>W</i> = {:.1f}; <i>p</i> {}'.format(self.statistic, fmt_p_html(self.pvalue))
	
	def summary(self):
		return 'W = {:.1f}; p {}'.format(self.statistic, fmt_p_text(self.pvalue))

def mannwhitney(df, dep, ind, *, nan_policy='warn', brunnermunzel=True, use_continuity=False, alternative='two-sided', method='auto'):
	"""
	Perform a Mann-Whitney test
	
	brunnermunzel: Set to False to skip the Brunner-Munzel test
	use_continuity, alternative, method: See scipy.stats.mannwhitneyu
	"""
	
	# Check for/clean NaNs
	df = check_nan(df[[ind, dep]], nan_policy)
	
	# Ensure 2 groups for ind
	group1, data1, group2, data2 = as_2groups(df, dep, ind)
	
	# Do Mann-Whitney test
	# Stata does not perform continuity correction
	result = stats.mannwhitneyu(data1, data2, use_continuity=use_continuity, alternative=alternative, method=method)
	
	u1 = result.statistic
	u2 = len(data1) * len(data2) - u1
	r = abs(2*u1 / (len(data1) * len(data2)) - 1)  # rank-biserial
	
	# If significant, perform a Brunner-Munzel test for our interest
	if result.pvalue < 0.05 and brunnermunzel:
		result_bm = stats.brunnermunzel(data1, data2)
		
		if result_bm.pvalue >= 0.05:
			warnings.warn('Mann-Whitney test is significant but Brunner-Munzel test is not. This could be due to a difference in shape, rather than location.')
			
			return MannWhitneyResult(
				statistic=min(u1, u2), pvalue=result.pvalue,
				#med1=data1.median(), med2=data2.median(),
				rank_biserial=r, direction=('{1} > {0}' if u1 < u2 else '{0} > {1}').format(group1, group2),
				brunnermunzel=BrunnerMunzelResult(statistic=result_bm.statistic, pvalue=result_bm.pvalue))
	
	return MannWhitneyResult(
		statistic=min(u1, u2), pvalue=result.pvalue,
		#med1=data1.median(), med2=data2.median(),
		rank_biserial=r, direction=('{1} > {0}' if u1 < u2 else '{0} > {1}').format(group1, group2))