def build_partial_corr(corr_df, target, covar, method='pearson', padjust='fdr_bh', pval=0.05, covar_name=None):
"""
Builds partial correlation DataFrame from corr_df of the target survey, controlling for covar.
corr_df (pd.DataFrame): correlation frame, assuming each row is an observation
target (str): targe column, can be a string prefix or suffix
covar (list): a list of covariates to control for
covar_name (str): optional name for covariates in the display
"""
partial_corr = pg.pairwise_corr(data=corr_df, covar=covar, method=method)
_, p_adj = pg.multicomp(partial_corr['p-unc'].values, alpha=pval, method=padjust)
partial_corr['p-corr'] = p_adj
partial_corr = partial_corr.loc[(partial_corr['p-corr'] < pval) & (~partial_corr['X'].str.contains(target)) & (partial_corr['Y'].str.contains(target))]
partial_corr['r_ctl'] = partial_corr['r']
partial_corr['p_ctl'] = partial_corr['p-corr']
if covar_name is not None:
partial_corr['covar'] = covar_name
partial_corr = partial_corr[['X', 'Y', 'covar', 'r_ctl', 'p_ctl']]
# drop the controlling covars for the raw pairwise correlation
pairwise_corr = pg.pairwise_corr(data=corr_df.drop(covar, axis='columns'), method=method, padjust=padjust)
pairwise_corr['r_unctl'] = pairwise_corr['r']
pairwise_corr['p_unctl'] = pairwise_corr['p-corr']
partial_corr = partial_corr.merge(pairwise_corr[['X', 'Y', 'r_unctl', 'p_unctl', 'n']], on=['X', 'Y'], how='left').sort_values('p_ctl')
return partial_corr.style.set_caption(method)
# add stats to title
for ax in grid.axes.flat:
ax.set(yscale="symlog")
ax.set_ylim(bottom=0)
var = ax.get_title().replace("population = ", "")
try:
child, parent = re.findall(r"(.*)/(.*)", var)[0]
ax.set_title(child)
ax.set_ylabel("Cells / uL")
except IndexError:
ax.set_title(var)
grid.savefig(figfile)
plt.close(grid.fig)
import pingouin as pg
m = matrix.join(meta[["severity_group"]])
m["severity_group"] = m["severity_group"].cat.remove_unused_categories()
res = pd.concat(
[
pg.pairwise_ttests(
data=m, dv=var, between="severity_group", parametric=False
).assign(variable=var)
for var in m.columns[:-1]
]
).drop(["Contrast"], axis=1)
res["p-cor"] = pg.multicomp(res["p-unc"].values, method="fdr_bh")[1]
res.to_csv("diff.absolute.csv", index=False)
plt.hist(x=dataFrame.slope)
#plt.show()
if withImage_anova == 'yes':
# 2-way ANOVA
aov = pg.mixed_anova(dv='slope',
between='group',
within='eye',
subject='subject',
data=dataFrame)
aov.round(3)
aov
# Bonferroni correction
pvals = [aov['p-unc'][0], aov['p-unc'][1], aov['p-unc'][2]]
reject, pvals_corr = pg.multicomp(pvals, method='fdr_bh')
print(reject, pvals_corr)
for sub in subjects:
data = cs_data.loc[cs_data.id == sub]
y, slope, eye = getRegressionCoeff(data)
x = data.test_num.unique().tolist()
# Assign color for plots
if sub[0:3] == 'ASW':
plot_palette = plot_colors[3]
elif sub[0:2] == 'AS' or sub[0:2] == 'AM':
plot_palette = plot_colors[0]
elif sub[0:2] == 'AA':
def stats(model, quantity, data, targets, tw, rm, nd):
if model == 'absolute':
data = data.drop(['NormQuant'], axis=1)
data['NormMean'] = data['NormMean'].astype(float)
mean = 'NormMean'
else:
data = data.drop(['rq'], axis=1)
data['rqMean'] = data['rqMean'].astype(float)
mean = 'rqMean'
# prepare data from intermediate dataframe
data = data[data['Outliers'].eq(False)]
data = data.drop_duplicates(keep='first')
# t-test and anova for normally distributed data
if nd == 'True':
if quantity == 2:
# T-Test between 2 groups
stats_dfs = pandas.DataFrame()
posthoc_dfs = pandas.DataFrame()
group = data['Group'].dropna()
group = group.drop_duplicates(keep='first').values.tolist()
for item in targets:
df = data[data['Target Name'].eq(item)]
group1 = df[df['Group'].eq(group[0])][mean]
group2 = df[df['Group'].eq(group[1])][mean]
t_test = ttest(group1, group2, paired=bool(rm))
if rm == 'True':
t_test['paired'] = 'TRUE'
else:
t_test['paired'] = 'FALSE'
t_test['Target Name'] = item
if stats_dfs is None:
stats_dfs = t_test
else:
stats_dfs = stats_dfs.append(t_test, ignore_index=True)
# reformat output table
stats_dfs = stats_dfs.rename(columns={
'cohen-d': 'effect size',
'BF10': 'Bayes factor',
'dof': 'DF'
})
cols = [
'Target Name', 'DF', 'T', 'tail', 'paired', 'p-val',
'effect size', 'power', 'Bayes factor'
]
stats_dfs = stats_dfs.reindex(columns=cols)
elif quantity >= 3:
# ANOVA test
stats_dfs = pandas.DataFrame()
posthoc_dfs = pandas.DataFrame()
# tukey_dfs = pandas.DataFrame()
pvals = []
for item in targets:
if rm == 'True':
# one-way
if tw == 'False':
# repeated measure anova
aov = pg.rm_anova(
dv=mean,
data=data[data['Target Name'].eq(item)],
within='Group',
subject='Sample Name',
detailed=True)
pvals.append(aov['p-unc'][0])
aov = aov.drop([1])
aov['measures'] = ['dependent']
aov['Target Name'] = item
# two-way
else:
aov = pg.rm_anova(
dv=mean,
data=data[data['Target Name'].eq(item)],
within=['Group1', 'Group2'],
subject='Sample Name',
detailed=True)
reject_tw, pval_corr_tw = pg.multicomp(list(
aov['p-unc']),
alpha=0.05,
method='bonf')
aov['p-value corrected'] = pval_corr_tw
aov['measures'] = ['dependent'] * 3
aov['Target Name'] = [item] * 3
aov.drop(['eps'], axis=1)
ph = pairwise_ttests(
data=data[data['Target Name'].eq(item)],
dv=mean,
within='Group',
subject='Sample Name',
padjust='fdr_bh')
ph['Target Name'] = item
ph['Test'] = 'T-Test'
else:
# one-way
if tw == 'False':
aov = pg.anova(dv=mean,
between='Group',
data=data[data['Target Name'].eq(item)],
detailed=True)
pvals.append(aov['p-unc'][0])
aov = aov.drop([1])
aov['measures'] = ['independent']
aov['Target Name'] = item
ph = pairwise_ttests(
data=data[data['Target Name'].eq(item)],
dv=mean,
between='Group',
padjust='fdr_bh')
ph['Test'] = 'T-Test'
# two-way
else:
aov = pg.anova(dv=mean,
between=['Group1', 'Group2'],
data=data[data['Target Name'].eq(item)],
detailed=False)
aov = aov.drop([3])
reject_tw, pval_corr_tw = pg.multicomp(list(
aov['p-unc']),
alpha=0.05,
method='bonf')
aov['p-value corrected'] = pval_corr_tw
aov['measures'] = ['independent'] * 3
aov['Target Name'] = [item] * 3
ph = pairwise_ttests(
data=data[data['Target Name'].eq(item)],
dv=mean,
between=['Group1', 'Group2'],
padjust='fdr_bh')
ph['Test'] = 'T-Test'
ph['Target Name'] = item
if stats_dfs is None:
stats_dfs = aov
else:
stats_dfs = stats_dfs.append(aov, ignore_index=True)
if posthoc_dfs is None:
posthoc_dfs = ph
else:
posthoc_dfs = posthoc_dfs.append(ph, ignore_index=True)
reject, pvals_corr = pg.multicomp(pvals, alpha=0.05, method='bonf')
# reformat output tables
stats_dfs = stats_dfs.rename(columns={
'p-unc': 'p-value',
'np2': 'effect size'
})
if tw == 'False':
stats_dfs['p-value corrected'] = pvals_corr
stats_dfs['distribution'] = ['parametric'] * len(targets)
stats_dfs['test'] = ['ANOVA'] * len(targets)
stats_dfs['statistic'] = ['NA'] * len(targets)
else:
stats_dfs['distribution'] = ['parametric'] * (len(targets) * 3)
stats_dfs['test'] = ['ANOVA'] * (len(targets) * 3)
stats_dfs['statistic'] = ['NA'] * (len(targets) * 3)
cols = [
'Target Name', 'Source', 'DF', 'F', 'MS', 'SS', 'p-value',
'p-value corrected', 'measures', 'distribution', 'test',
'statistic', 'effect size'
]
stats_dfs = stats_dfs.reindex(columns=cols)
if tw == 'False':
posthoc_dfs = posthoc_dfs.drop(['Contrast', 'T'], axis=1)
else:
posthoc_dfs = posthoc_dfs.drop(['T'], axis=1)
posthoc_dfs = posthoc_dfs.rename(
columns={
'hedges': 'effect size',
'p-corr': 'p-value corrected',
'p-unc': 'p-value',
'p-adjust': 'correction method',
'BF10': 'Bayes factor',
'dof': 'DF'
})
if tw == 'False':
cols2 = [
'Target Name', 'A', 'B', 'DF', 'p-value corrected',
'p-value', 'correction method', 'Paired', 'Parametric',
'Test', 'effect size', 'Bayes factor'
]
else:
cols2 = [
'Target Name', 'Contrast', 'Group1', 'A', 'B', 'DF',
'p-value corrected', 'p-value', 'correction method',
'Paired', 'Parametric', 'Test', 'effect size',
'Bayes factor'
]
posthoc_dfs = posthoc_dfs.reindex(columns=cols2)
# nonparametric tests for not normally distributed data
else:
if quantity == 2:
stats_dfs = pandas.DataFrame()
posthoc_dfs = pandas.DataFrame()
group = data['Group'].dropna()
group = group.drop_duplicates(keep='first').values.tolist()
for item in targets:
df = data[data['Target Name'].eq(item)]
group1 = df[df['Group'].eq(group[0])][mean]
group2 = df[df['Group'].eq(group[1])][mean]
if rm == 'True':
# Mann-Whitney U test
test = mannwhitneyu(group1, group2)
test = pandas.DataFrame(
{
'Target Name': item,
'pvalue': test.pvalue,
'statistic': test.statistic
},
index=[0])
else:
# Wilcoxon
test = wilcoxon(group1, group2)
test = pandas.DataFrame(
{
'Target Name': item,
'pvalue': test.pvalue,
'statistic': test.statistic
},
index=[0])
if stats_dfs is None:
stats_dfs = test
else:
stats_dfs = stats_dfs.append(test, ignore_index=True)
elif quantity >= 3:
stats_dfs = pandas.DataFrame()
posthoc_dfs = pandas.DataFrame()
pvals = []
for item in targets:
if rm == 'True':
# friedman test for repeated measurements
df = pg.friedman(dv=mean,
within='Group',
subject='Sample Name',
data=data[data['Target Name'].eq(item)])
pvals.append(df['p-unc'][0])
df['test'] = ['Friedman Q']
df['measures'] = ['dependent']
df = df.rename(columns={'Q': 'statistic'})
df['Target Name'] = item
df['DF'] = 'NA'
ph = pairwise_ttests(
data=data[data['Target Name'].eq(item)],
dv=mean,
within='Group',
subject='Sample Name',
padjust='fdr_bh',
parametric=False)
ph['Target Name'] = item
ph['DF'] = 'NA'
ph['Bayes factor'] = 'NA'
ph['Test'] = 'Wilcoxon'
else:
# Kruskal-Wallis H test
df = pg.kruskal(dv=mean,
between='Group',
data=data[data['Target Name'].eq(item)])
pvals.append(df['p-unc'][0])
df['test'] = ['Kruskal-Wallis H']
df['measures'] = ['independent']
df = df.rename(columns={'H': 'statistic'})
df['Target Name'] = item
df['DF'] = 'NA'
ph = pairwise_ttests(
data=data[data['Target Name'].eq(item)],
dv=mean,
between='Group',
padjust='fdr_bh',
parametric=False)
ph['Target Name'] = item
ph['DF'] = 'NA'
ph['Bayes factor'] = 'NA'
ph['Test'] = 'Mann-Whitney U'
if stats_dfs is None:
stats_dfs = df
else:
stats_dfs = stats_dfs.append(df, ignore_index=True)
if posthoc_dfs is None:
posthoc_dfs = ph
else:
posthoc_dfs = posthoc_dfs.append(ph, ignore_index=True)
reject, pvals_corr = pg.multicomp(pvals, alpha=0.05, method='bonf')
# reformat output tables
stats_dfs = stats_dfs.rename(columns={
'dof': 'DF',
'p-unc': 'p-value'
})
stats_dfs['p-value corrected'] = pvals_corr
stats_dfs['distribution'] = ['non-parametric'] * len(targets)
stats_dfs['MS'] = ['NA'] * len(targets)
stats_dfs['SS'] = ['NA'] * len(targets)
stats_dfs['effect size'] = ['NA'] * len(targets)
cols = [
'Target Name', 'DF', 'MS', 'SS', 'p-value',
'p-value corrected', 'measures', 'distribution', 'test',
'statistic', 'effect size'
]
stats_dfs = stats_dfs.reindex(columns=cols)
posthoc_dfs = posthoc_dfs.drop(['Contrast'], axis=1)
posthoc_dfs = posthoc_dfs.rename(
columns={
'hedges': 'effect size',
'p-corr': 'p-value corrected',
'p-unc': 'p-value',
'p-adjust': 'correction method',
'BF10': 'Bayes factor'
})
cols2 = [
'Target Name', 'A', 'B', 'DF', 'p-value corrected', 'p-value',
'correction method', 'Paired', 'Parametric', 'Test',
'effect size', 'Bayes factor'
]
posthoc_dfs = posthoc_dfs.reindex(columns=cols2)
return stats_dfs, posthoc_dfs
def check_mcnemar_significance(mcnemar_pvals):
import pingouin as pg
reject, pvals = pg.multicomp(mcnemar_pvals, alpha=0.05, method="holm")
return reject, pvals
def swarmboxenplot(
data: DataFrame,
x: str,
y: tp.Union[str, Iterables],
hue: str = None,
swarm: bool = True,
boxen: bool = True,
bar: bool = False,
ax: tp.Union[Axis, tp.Sequence[Axis]] = None,
test: tp.Union[bool, str] = "mann-whitney",
multiple_testing: tp.Union[bool, str] = "fdr_bh",
test_upper_threshold: float = 0.05,
test_lower_threshold: float = 0.01,
plot_non_significant: bool = False,
plot_kws: tp.Dict[str, tp.Any] = None,
test_kws: tp.Dict[str, tp.Any] = None,
fig_kws: tp.Dict[str, tp.Any] = None,
) -> tp.Optional[tp.Union[Figure, DataFrame, tp.Tuple[Figure, DataFrame]]]:
"""
A categorical plot that overlays individual observations
as a swarm plot and summary statistics about them in a boxen plot.
In addition, this plot will test differences between observation
groups and add lines representing a significant difference between
them.
Parameters
----------
data: pd.DataFrame
A dataframe with data where the rows are the observations and
columns are the variables to group them by.
x: str
The categorical variable.
y: str | list[str]
The continuous variable to plot.
If more than one is given, will ignore the `ax` attribute and
return figure with a subplot per each `y` variable.
hue: str, optional
An optional categorical variable to further group observations by.
swarm: bool
Whether to plot individual observations as a swarmplot.
boxen: bool
Whether to plot summary statistics as a boxenplot.
ax: matplotlib.axes.Axes, optional
An optional axes to draw in.
test: bool | str
Whether to test differences between observation groups.
If `False`, will not return a dataframe as well.
If a string is passed, will perform test accordingly. Available tests:
- 't-test':
- 'mann-whitney':
- 'kruskal':
Default is a parwise 'mann-whitney' test with p-value adjustment.
multiple_testing: str
Method for multiple testing correction.
test_upper_threshold: float
Upper theshold to consider p-values significant.
Will be marked with "*".
test_lower_threshold: float
Secondary theshold to consider p-values highly significant.
Will be marked with "**".
plot_non_significant: bool
Whether to add a "n.s." sign to p-values above `test_upper_threshold`.
plot_kws: dict
Additional values to pass to seaborn.boxenplot or seaborn.swarmplot
test_kws: dict
Additional values to pass to pingouin.pairwise_ttests.
The default is: dict(parametric=False) to run a non-parametric test.
Returns
-------
tuple[Figure, pandas.DataFrame]:
if `ax` is None and `test` is True.
pandas.DataFrame: if `ax` is not None.
Figure: if `test` is False.
None:
if `test` is False and `ax` is not None.
Raises
------
ValueError:
If either the `x` or `hue` column in `data` are not
Category, string or object type, or if `y` is not numeric.
"""
# opts = dict(data=data, x='h', y='y', hue='x', test_kws=dict(parametric=False))
# opts = dict(data=data, x='cat', y='cont')
# for k, v in opts.items():
# locals()[k] = v
for var, name in [(x, "x"), (hue, "hue")]:
if var is not None:
if not data[var].dtype.name in ["category", "string", "object"]:
raise ValueError(
f"`{name}` variable must be categorical, string or object."
)
if test_kws is None:
test_kws = dict()
if plot_kws is None:
plot_kws = dict()
data = data.sort_values([x] + ([hue] if hue is not None else []))
if isinstance(y, (list, pd.Series, pd.Index)):
# TODO: display only one legend for hue
if ax is None:
n, m = get_grid_dims(y)
default_fig_kws = dict(nrows=n,
ncols=m,
figsize=(m * 4, n * 4),
sharex=True,
squeeze=False)
default_fig_kws.update(fig_kws or {})
fig, axes = plt.subplots(**default_fig_kws)
axes = axes.flatten()
elif isinstance(ax, np.ndarray):
axes = ax.flatten()
elif isinstance(ax, matplotlib.axes.Axes):
axes = np.asarray([ax])
_stats = list()
idx = -1
for idx, _var in enumerate(y):
_ax = axes[idx]
s: DataFrame = swarmboxenplot(
data=data,
x=x,
y=_var,
hue=hue,
swarm=swarm,
boxen=boxen,
bar=bar,
ax=_ax,
test=test,
multiple_testing=multiple_testing,
test_upper_threshold=test_upper_threshold,
test_lower_threshold=test_lower_threshold,
plot_non_significant=plot_non_significant,
plot_kws=plot_kws,
test_kws=test_kws,
)
_ax.set(title=_var + _ax.get_title(), xlabel=None, ylabel=None)
if test is not False:
_stats.append(s.assign(Variable=_var))
# "close" excess subplots
for _ax in axes[idx + 1:]:
_ax.axis("off")
if test is not False:
stats = pd.concat(_stats).reset_index(drop=True)
cols = [c for c in stats.columns if c != "Variable"]
stats = stats.reindex(["Variable"] + cols, axis=1)
# If there is just one test per `y` (no hue), correct p-values
if stats.shape == len(y):
stats["p-cor"] = pg.multicomp(stats["p-unc"].values,
method=multiple_testing)[1]
if ax is None:
return (fig, stats) if test else fig
return stats if test else None
if data[y].dtype.name in ["category", "string", "object"]:
raise ValueError("`y` variable must be numeric.")
if ax is None:
fig, _ax = plt.subplots(1, 1, figsize=(4, 4))
else:
_ax = ax
# Plot vanilla seaborn
if boxen:
assert not bar
# Tmp fix for lack of support for Pandas Int64 in boxenplot:
if data[y].dtype.name == "Int64":
data[y] = data[y].astype(float)
boxen_kws = filter_kwargs_by_callable(plot_kws, sns.boxenplot)
sns.boxenplot(data=data, x=x, y=y, hue=hue, ax=_ax, **boxen_kws)
if bar:
assert not boxen
bar_kws = filter_kwargs_by_callable(plot_kws, sns.barplot)
sns.barplot(data=data, x=x, y=y, hue=hue, ax=_ax, **bar_kws)
if (boxen or bar) and swarm:
_add_transparency_to_plot(_ax, kind="bar" if bar else "boxen")
if swarm:
swarm_kws = filter_kwargs_by_callable(plot_kws, sns.swarmplot)
if hue is not None and "dodge" not in swarm_kws:
swarm_kws["dodge"] = True
with warnings.catch_warnings():
warnings.filterwarnings("ignore", category=UserWarning)
sns.swarmplot(data=data, x=x, y=y, hue=hue, ax=_ax, **swarm_kws)
_ax.set_xticklabels(_ax.get_xticklabels(), rotation=90, ha="right")
if test is not False:
if test in [True, "t-test", "mann-whitney"]:
test_function = pg.pairwise_ttests
if test == "mann-whitney":
test_kws["parametric"] = False
elif test in ["kruskal"]:
test_function = pg.kruskal
assert hue is None, "If test is 'kruskal', 'hue' must be None."
else:
raise ValueError(f"Test type '{test}' not recognized.")
#
if not data.index.is_unique:
print("Warning: dataframe contains a duplicated index.")
# remove NaNs
datat = data.dropna(subset=[x, y] + ([hue] if hue is not None else []))
# remove categories with only one element
keep = datat.groupby(x).size()[datat.groupby(x).size() > 1].index
datat = datat.loc[datat[x].isin(keep), :]
if datat[x].dtype.name == "category":
datat[x] = datat[x].cat.remove_unused_categories()
ylim = _ax.get_ylim() # save original axis boundaries for later
ylength = abs(ylim[1]) + (abs(ylim[0]) if ylim[0] < 0 else 0)
# Now calculate stats
# # get empty dataframe in case nothing can be calculated
stat = _get_empty_stat_results(datat, x, y, hue, add_median=True)
# # mirror groups to account for own pingouin order
tats = stat.rename(columns={
"B": "A",
"A": "B",
"median_A": "median_B",
"median_B": "median_A",
})
stat = (pd.concat([stat,
tats]).sort_values(["Contrast", "A",
"B"]).reset_index(drop=True))
try:
_stat = test_function(
data=datat,
dv=y,
between=x if hue is None else [x, hue],
**test_kws,
)
except (AssertionError, ValueError) as e:
print(str(e))
_stat = stat
except KeyError:
print("Only one category with values!")
_stat = stat
if test == "kruskal":
p = _stat.squeeze()["p-unc"]
symbol = ("**" if p <= test_lower_threshold else "n.s." if
((p > test_upper_threshold) or pd.isnull(p)) else "*")
_ax.set_title(symbol)
return (fig, _stat) if ax is None else _stat
stat = _stat.merge(
stat[["Contrast", "A", "B", "median_A", "median_B"] +
([x] if hue is not None else [])],
how="left",
)
if multiple_testing is not False:
if "p-unc" not in stat.columns:
stat["p-unc"] = np.nan
stat["p-cor"] = pg.multicomp(stat["p-unc"].values,
method=multiple_testing)[1]
pcol = "p-cor"
else:
pcol = "p-unc"
# This ensures there is a point for each `x` class and keep the order
# correct for below
mm = data.groupby([x] + ([hue] if hue is not None else []))[y].median()
if hue is None:
order = {k: float(i) for i, k in enumerate(mm.index)}
else:
nhues = data[hue].drop_duplicates().dropna().shape[0]
order = {
k: (float(i) / nhues) - (1 / nhues) - 0.05
for i, k in enumerate(mm.index)
}
_ax.scatter(order.values(), mm, alpha=0, color="white")
# Plot significance bars
# start at top of the plot and progressively decrease sig. bar downwards
py = data[y].max()
incr = ylength / 100 # divide yaxis in 100 steps
for idx, row in stat.iterrows():
p = row[pcol]
if (pd.isnull(p) or
(p > test_upper_threshold)) and (not plot_non_significant):
py -= incr
continue
symbol = ("**" if p <= test_lower_threshold else "n.s." if
((p > test_upper_threshold) or pd.isnull(p)) else "*")
if hue is not None:
if row[x] != "-":
xx = (order[(row[x], row["A"])], order[(row[x], row["B"])])
else:
try:
# TODO: get more accurate middle of group
xx = (
order[(row["A"], stat["A"].iloc[-1])] -
(1 / nhues),
order[(row["B"], stat["B"].iloc[-1])] -
(1 / nhues),
)
except KeyError:
# These are the hue groups without contrasting on 'x'
continue
else:
xx = (order[row["A"]], order[row["B"]])
red_fact = 0.95 # make the end position shorter
_ax.plot(
(xx[0], xx[1] * red_fact),
(py, py),
color="black",
linewidth=1.2,
)
_ax.text(xx[1] * red_fact,
py,
s=symbol,
color="black",
ha="center")
py -= incr
_ax.set_ylim(ylim)
return (fig, stat) if ax is None else stat
return fig if ax is None else None
def qualOrdinalUnpaired(imgDir,
sheetName,
sheetDf,
sheetScale,
silent=False):
print("######################################## ", sheetName,
" ########################################"
) if not silent else None
meltedSheetDf = sheetDf.melt(var_name='Factor', value_name='Variable')
contingencySheetDf = pd.crosstab(index=meltedSheetDf['Variable'],
columns=meltedSheetDf['Factor'])
statDf = pd.DataFrame(columns=[
'COMPARISON', 'TEST', 'STATISTICS', 'P-VALUE', 'EFFECT SIZE'
])
#fill empty scale value
for sheetStep in range(sheetScale):
if not sheetStep in contingencySheetDf.index.values:
contingencySheetDf.loc[sheetStep] = [
0 for x in range(len(contingencySheetDf.columns.values))
]
contingencySheetDf.sort_index(inplace=True)
# ALL MODALITY
if len(contingencySheetDf.columns) > 2:
sheetDf_long = sheetDf.melt(ignore_index=False).reset_index()
kruskal_stats = pg.kruskal(data=sheetDf_long,
dv="value",
between="variable")
source, ddof1, hvalue, pvalue = kruskal_stats.values[0]
statDf = statDf.append(
{
'COMPARISON': 'ALL',
'TEST': "Kruskal-Wallis",
'STATISTICS': hvalue,
'P-VALUE': pvalue,
'EFFECT SIZE': -1
},
ignore_index=True)
# BETWEEN MODALITY
modality_names = sheetDf.columns.values
uncorrectedStatIndex = len(statDf.index)
for i in range(len(modality_names)):
for j in range(i + 1, len(modality_names)):
stats_mannwhitney = pg.mwu(x=sheetDf.loc[:, modality_names[i]],
y=sheetDf.loc[:, modality_names[j]],
alternative='two-sided')
uvalue, alternative, pvalue, RBC, CLES = stats_mannwhitney.values[
0]
statDf = statDf.append(
{
'COMPARISON':
modality_names[i] + '|' + modality_names[j],
'TEST': "Mann-Whitney",
'STATISTICS': uvalue,
'P-VALUE': pvalue,
'EFFECT SIZE': RBC
},
ignore_index=True)
reject, statDf.loc[uncorrectedStatIndex::, 'P-VALUE'] = pg.multicomp(
statDf.loc[uncorrectedStatIndex::, 'P-VALUE'].values,
alpha=0.05,
method="holm")
StackedBarPlotter.StackedBarPlotter(filename=imgDir + '/' + sheetName +
'.png',
title=sheetName,
dataDf=sheetDf,
histDf=contingencySheetDf,
statDf=statDf)