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)
def display_all_corr(corr_df, cols, target, alpha=0.1, method="pearson"):
"""Displays all correlations for every feature in cols against target.
Sorts by feature group.
Also stars the significant p-values.
Args:
corr_df (pd.df): pandas correlation frame
cols (list): list of feature columns
target (str): the target survey column name
alpha (float): the specified significance level
method (str): whether "pearson" or "spearman" correlations
Returns:
pd.df: styled correlation matrix
"""
p_val_format = "{:0.4f}"
style_p_val = lambda x: "{:0.4f}".format(x) + "*" if x < alpha else "{:0.4f}".format(x)
pair_corr_df = pg.pairwise_corr(corr_df, columns=[cols, [target]], method=method, padjust="fdr_bh")
pair_corr_df['p-corr'] = pair_corr_df['p-corr'].apply(style_p_val)
pair_corr_df['p-unc'] = pair_corr_df['p-unc'].apply(style_p_val)
title = method + ", alpha < {}".format(alpha)
display(pair_corr_df[['X', 'Y', 'n', 'r', 'p-unc', 'p-corr', 'p-adjust']].style.set_caption(title))
def build_corr_table(df, x, cols, target, title, method='pearson', groups=['all', 'no_symp', 'soc_anx', 'gen_anx', 'dep_anx']):
"""Builds a concise correlation table for the given feature target across all subgroups."""
table = pd.DataFrame()
for group in groups:
if group == 'all':
sel_df = df
else:
sel_df = df[df['baseline_cluster'] == group]
corr_df = pg.pairwise_corr(sel_df, columns=[cols, [target]], nan_policy='pairwise', padjust="fdr_bh", method=method)
corr_df['group'] = group
#print(corr_df.head())
corr_df = corr_df[corr_df['X'] == x]
table = table.append(corr_df[['group', 'n', 'r', 'p-corr', 'p-adjust']])
table = table.reset_index(drop=True)
return table.style.set_caption(title)
def run_pheno_partialcorrs(df_phenos, df_z, method='pearson'):
df_input = pd.concat((df_phenos, df_z), axis=1)
if method == 'pearson':
df_out = pd.DataFrame(
columns=['pheno', 'variable', 'coef', 'p', 'BF10'])
else:
df_out = pd.DataFrame(columns=['pheno', 'variable', 'coef', 'p'])
phenos = list(df_phenos.columns)
for pheno in phenos:
print(pheno)
if method == 'pearson':
df_tmp = pd.DataFrame(index=df_z.columns,
columns=['coef', 'p', 'BF10'])
else:
df_tmp = pd.DataFrame(index=df_z.columns, columns=['coef', 'p'])
phenos_cov = phenos.copy()
phenos_cov.remove(pheno)
results = pg.pairwise_corr(data=df_input,
columns=[[pheno],
list(df_z.columns)],
covar=phenos_cov,
method=method)
results.set_index('Y', inplace=True)
df_tmp.loc[:, 'coef'] = results['r']
df_tmp.loc[:, 'p'] = results['p-unc']
if method == 'pearson':
df_tmp.loc[:, 'BF10'] = results['BF10'].astype(float)
# append
df_tmp.reset_index(inplace=True)
df_tmp.rename(index=str, columns={'index': 'variable'}, inplace=True)
df_tmp['pheno'] = pheno
df_out = df_out.append(df_tmp, sort=False)
df_out.set_index(['pheno', 'variable'], inplace=True)
return df_out
def test_pandas(self):
"""Test pandas method.
"""
# Test the ANOVA (Pandas)
aov = df.anova(dv='Scores', between='Group', detailed=True)
assert aov.equals(
pg.anova(dv='Scores', between='Group', detailed=True, data=df))
aov3_ss1 = df_aov3.anova(dv='Cholesterol',
between=['Sex', 'Drug'],
ss_type=1)
aov3_ss2 = df_aov3.anova(dv='Cholesterol',
between=['Sex', 'Drug'],
ss_type=2)
aov3_ss2_pg = pg.anova(dv='Cholesterol',
between=['Sex', 'Drug'],
data=df_aov3,
ss_type=2)
assert not aov3_ss1.equals(aov3_ss2)
assert aov3_ss2.round(3).equals(aov3_ss2_pg.round(3))
# Test the Welch ANOVA (Pandas)
aov = df.welch_anova(dv='Scores', between='Group')
assert aov.equals(pg.welch_anova(dv='Scores', between='Group',
data=df))
# Test the ANCOVA
aov = df_anc.ancova(dv='Scores', covar='Income',
between='Method').round(3)
assert (aov.equals(
pg.ancova(data=df_anc,
dv='Scores',
covar='Income',
between='Method').round(3)))
# Test the repeated measures ANOVA (Pandas)
aov = df.rm_anova(dv='Scores',
within='Time',
subject='Subject',
detailed=True)
assert (aov.equals(
pg.rm_anova(dv='Scores',
within='Time',
subject='Subject',
detailed=True,
data=df)))
# FDR-corrected post hocs with Hedges'g effect size
ttests = df.pairwise_tests(dv='Scores',
within='Time',
subject='Subject',
padjust='fdr_bh',
effsize='hedges')
assert (ttests.equals(
pg.pairwise_tests(dv='Scores',
within='Time',
subject='Subject',
padjust='fdr_bh',
effsize='hedges',
data=df)))
# Pairwise Tukey
tukey = df.pairwise_tukey(dv='Scores', between='Group')
assert tukey.equals(
pg.pairwise_tukey(data=df, dv='Scores', between='Group'))
# Test two-way mixed ANOVA
aov = df.mixed_anova(dv='Scores',
between='Group',
within='Time',
subject='Subject',
correction=False)
assert (aov.equals(
pg.mixed_anova(dv='Scores',
between='Group',
within='Time',
subject='Subject',
correction=False,
data=df)))
# Test parwise correlations
corrs = data.pairwise_corr(columns=['X', 'M', 'Y'], method='spearman')
corrs2 = pg.pairwise_corr(data=data,
columns=['X', 'M', 'Y'],
method='spearman')
assert corrs['r'].equals(corrs2['r'])
# Test partial correlation
corrs = data.partial_corr(x='X', y='Y', covar='M', method='spearman')
corrs2 = pg.partial_corr(x='X',
y='Y',
covar='M',
method='spearman',
data=data)
assert corrs['r'].equals(corrs2['r'])
# Test partial correlation matrix (compare with the ppcor package)
corrs = data.iloc[:, :5].pcorr().round(3)
np.testing.assert_array_equal(corrs.iloc[0, :].to_numpy(),
[1, 0.392, 0.06, -0.014, -0.149])
# Now compare against Pingouin's own partial_corr function
corrs = data[['X', 'Y', 'M']].pcorr()
corrs2 = data.partial_corr(x='X', y='Y', covar='M')
assert np.isclose(corrs.at['X', 'Y'], corrs2.at['pearson', 'r'])
# Test rcorr (correlation matrix with p-values)
# We compare against Pingouin pairwise_corr function
corrs = df_corr.rcorr(padjust='holm', decimals=4)
corrs2 = df_corr.pairwise_corr(padjust='holm').round(4)
assert corrs.at['Neuroticism', 'Agreeableness'] == '*'
assert (corrs.at['Agreeableness',
'Neuroticism'] == str(corrs2.at[2, 'r']))
corrs = df_corr.rcorr(padjust='holm', stars=False, decimals=4)
assert (corrs.at['Neuroticism',
'Agreeableness'] == str(corrs2.at[2,
'p-corr'].round(4)))
corrs = df_corr.rcorr(upper='n', decimals=5)
corrs2 = df_corr.pairwise_corr().round(5)
assert corrs.at['Extraversion', 'Openness'] == corrs2.at[4, 'n']
assert corrs.at['Openness', 'Extraversion'] == str(corrs2.at[4, 'r'])
# Method = spearman does not work with Python 3.5 on Travis?
# Instead it seems to return the Pearson correlation!
df_corr.rcorr(method='spearman')
df_corr.rcorr()
# Test mediation analysis
med = data.mediation_analysis(x='X', m='M', y='Y', seed=42, n_boot=500)
np.testing.assert_array_equal(med.loc[:, 'coef'].round(4).to_numpy(),
[0.5610, 0.6542, 0.3961, 0.0396, 0.3565])
else:
print('Null hypothesis does not have enough evidence to be rejected.\n Failed to reject null hypothesis.')
"""$ p < 0.005 $
Since our p value was far lesser than the statistical significance value, we can reliably reject the null hypothesis, in favour of the alternative.
Additionally, we obtained a negative t-test statistic score, which implies that the mean of international students in UK universities is greater than in US universities.
We reject $H_o : \mu_x = \mu_y$ , where<br><br>$x :$Set of UK international_students<br>$y :$ Set of US international_students
### Correlation
"""
# Table for the pair-wise correlation of features x and y
corr_data = pg.pairwise_corr(data,columns=numerical_vars)
corr_data
"""**Analyze the correlation between `teaching` and `total_score`**"""
# Plot a scatter plot between teaching and total_score
top2011 = data.loc[data['year'] == 2011].head(100)
x = top2011['total_score']
y = top2011['teaching']
plt.figure(figsize=(5, 5))
ax = sns.scatterplot(x=x,
y=y)
"""Hence, we can conclude that universities with greater overall score have better teaching scores. The relation follows a roughly linear path with positive slope.
usecols="B:Q")
io_2050rcp8p5 = pd.read_excel('genie_outpout.xlsx',
sheet_name='io_STATS_foram_only',
usecols="T:AI")
io_2100rcp8p5 = pd.read_excel('genie_outpout.xlsx',
sheet_name='io_STATS_foram_only',
usecols="AL:BA")
io_2100rcp6 = pd.read_excel('\genie_outpout.xlsx',
sheet_name='io_STATS_foram_only',
usecols="BD:BS")
###########################################################################################################################
## spearman pairwise_corr, foram vs all
#subpolar
corr_subpolar_present = pd.DataFrame(
pg.pairwise_corr(subpolar_present, columns=['foram'], method='spearman'))
corr_subpolar_2050rcp8p5 = pd.DataFrame(
pg.pairwise_corr(subpolar_2050rcp8p5,
columns=['foram.1'],
method='spearman'))
corr_subpolar_2100rcp8p5 = pd.DataFrame(
pg.pairwise_corr(subpolar_2100rcp8p5,
columns=['foram.2'],
method='spearman'))
#corr_subpolar_2100rcp6 = pd.DataFrame(pg.pairwise_corr(subpolar_2100rcp6, columns=['foram.3'], method='spearman'))
#temperate
corr_temp_present = pd.DataFrame(
pg.pairwise_corr(temp_present, columns=['foram'], method='spearman'))
corr_temp_2050rcp8p5 = pd.DataFrame(
pg.pairwise_corr(temp_2050rcp8p5, columns=['foram.1'], method='spearman'))
corr_temp_2100rcp8p5 = pd.DataFrame(
x_pos = expand.slider("X position for the label", 0.0, max_x, start_x,
(max_x / 100 + 0.1))
y_pos = expand.slider("Y position for the label", 0.0, max_y, start_y,
(max_y / 100 + 0.1))
sns.set(style='white', font_scale=font_scale)
st.success("Correlation results")
corr_result = pg.corr(x=df[x_var1], y=df[x_var2], method=method_selected)
st.write(corr_result)
st.success("Correlation matrices")
st.write(
pg.pairwise_corr(df, padjust='bonf',
method=method_selected).sort_values(by=['p-unc']))
st.write(df.rcorr(padjust='bonf'))
st.success("Correlation plot with distributions is being generated")
fig = plt.figure(figsize=(12, 6))
g = sns.JointGrid(data=df, x=x_var1, y=x_var2, height=6)
g = g.plot_joint(sns.regplot, color="xkcd:muted blue")
g = g.plot_marginals(sns.distplot,
kde=False,
bins=12,
color="xkcd:bluey grey")
if plot_r:
g.ax_joint.text(
x_pos,
y_pos,
import pandas as pd
import numpy as np
import pingouin as pg
io = r'Lake_v2.xls'
data = pd.read_excel(io, sheet_name=2, usecols=[2, 3, 4])
data.head()
print(len(data))
for i in range(len(data)):
print(data.loc[i])
corr = pg.pairwise_corr(data, method='pearson')
spearman_corr = pg.pairwise_corr(data, method='spearman')
kendall_corr = pg.pairwise_corr(data, method='kendall')
bicor_corr = pg.pairwise_corr(data, method='bicor')
skipped_corr = pg.pairwise_corr(data, method='skipped')
corr = corr.append(spearman_corr)
corr = corr.append(kendall_corr)
corr = corr.append(bicor_corr)
corr = corr.append(skipped_corr)
corr.to_excel('correlation.xls')
def test_pandas(self):
"""Test pandas method.
"""
# Test the ANOVA (Pandas)
aov = df.anova(dv='Scores', between='Group', detailed=True)
assert aov.equals(
pg.anova(dv='Scores', between='Group', detailed=True, data=df))
# Test the Welch ANOVA (Pandas)
aov = df.welch_anova(dv='Scores', between='Group')
assert aov.equals(pg.welch_anova(dv='Scores', between='Group',
data=df))
# Test the repeated measures ANOVA (Pandas)
aov = df.rm_anova(dv='Scores',
within='Time',
subject='Subject',
detailed=True)
assert aov.equals(
pg.rm_anova(dv='Scores',
within='Time',
subject='Subject',
detailed=True,
data=df))
# FDR-corrected post hocs with Hedges'g effect size
ttests = df.pairwise_ttests(dv='Scores',
within='Time',
subject='Subject',
padjust='fdr_bh',
effsize='hedges')
assert ttests.equals(
pg.pairwise_ttests(dv='Scores',
within='Time',
subject='Subject',
padjust='fdr_bh',
effsize='hedges',
data=df))
# Test two-way mixed ANOVA
aov = df.mixed_anova(dv='Scores',
between='Group',
within='Time',
subject='Subject',
correction=False)
assert aov.equals(
pg.mixed_anova(dv='Scores',
between='Group',
within='Time',
subject='Subject',
correction=False,
data=df))
# Test parwise correlations
corrs = data.pairwise_corr(columns=['X', 'M', 'Y'], method='spearman')
corrs2 = pg.pairwise_corr(data=data,
columns=['X', 'M', 'Y'],
method='spearman')
assert corrs['r'].equals(corrs2['r'])
# Test partial correlation
corrs = data.partial_corr(x='X', y='Y', covar='M', method='spearman')
corrs2 = pg.partial_corr(x='X',
y='Y',
covar='M',
method='spearman',
data=data)
assert corrs['r'].equals(corrs2['r'])
# Test partial correlation matrix (compare with the ppcor package)
corrs = data.pcorr().round(3)
np.testing.assert_array_equal(corrs.iloc[0, :].values,
[1, 0.392, 0.06, -0.014, -0.149])
# Now compare against Pingouin's own partial_corr function
corrs = data[['X', 'Y', 'M']].pcorr()
corrs2 = data.partial_corr(x='X', y='Y', covar='M')
assert round(corrs.loc['X', 'Y'], 3) == corrs2.loc['pearson', 'r']
# Test mediation analysis
med = data.mediation_analysis(x='X', m='M', y='Y', seed=42, n_boot=500)
np.testing.assert_array_equal(med.loc[:, 'coef'].values,
[0.5610, 0.6542, 0.3961, 0.0396, 0.3565])
columns = list(df)
features = []
for column in df:
epiclist = df[
column] # epiclist = values for a certain feature for all iterations
features.append(epiclist)
#generates a heatmap that we can use to find strongly correlated values
plt.figure(figsize=(12, 10))
cor = df.corr(method='pearson')
print("checkpoint")
print(
pg.pairwise_corr(df).sort_values(by=['p-unc'])[[
'X', 'Y', 'n', 'r', 'p-unc'
]])
print("checkpoint 2")
print(cor)
pingouin_raw_corr = pg.pairwise_corr(df).sort_values(by=['p-unc'])[[
'X', 'Y', 'n', 'r', 'p-unc'
]]
pingouin_corr = pd.DataFrame()
for x in df.columns:
for y in df.columns:
df[x] = np.nan_to_num(df[x])
df[y] = np.nan_to_num(df[y])
def correlations(df, var_list, tail, correction, method, cov_list, *args):
"""
Compute correlations por each group of variables dropping variables of the same group (i.e. if
we have Depression_mother, Depression_father, Depression_daughter, correlations between these
variables that are part of the same set of variables are excluded, only if we specified a common
name of these variables ('Depression') in *args. Otherwise they are included.
Apply a correction for multiple comparisons for all uncorrected p-values of each group of variables,
not to all the correlations in the DataFrame. Joins everything in an unique DataFrame.
Input:
df: DataFame with data.
var_list: list of lists with variables for correlations.
tail (str): 'one-sided' or 'two-sided'.
correction (str): check statsmodels.multitest.multipletest for correction options.
method (str): 'pearson', 'spearman', etc. (check methods for pg.pairwise_corr).
cov_list: list of covariates.
*args (str): strings to exclude intra group correlations from DataFrame and apply
correction without these correlations. The string should be a name that
is common to all the names of the variables that are part of the group.
Specify one string for the X column of the DataFrame:
- 'Depression' if is the common string - Depression_mother, Depression_father, etc.
and we just have that group for the X column
- 'Depression|Anger' if we have two groups - Depression_mother, Depression_father, etc. +
Anger_mother, Anger_father, etc.
Specify another string with the same logic for the Y column of the DataFrame.
Output:
df: DataFrame with correlation values dropping correlations between variables of the same group.
"""
list_of_dfs = []
# list_to_exclude_both_rows = [excludeintra_X, excludeintra_Y, excludeintra_Z]
df_corr = pd.DataFrame(
pg.pairwise_corr(df,
columns=var_list,
covar=cov_list,
tail=tail,
method=method,
nan_policy='pairwise')).round(5)
# drop intra task correlations. Reset index because of dropped rows
# to ease concat
for row_excl in args:
# drop rows that contains the same string in both columns X and Y
df_corr = df_corr[~(
(df_corr['X'].str.contains(row_excl)) &
(df_corr['Y'].str.contains(row_excl)))].reset_index(drop=True)
# apply correction to p-values of no intratask correlations
# transpose df as it returns two rows and multiple columns
# and rename columns
FDR_corr = list(
sm.multitest.multipletests(df_corr['p-unc'],
alpha=0.05,
method=correction,
is_sorted=False,
returnsorted=False))
# extract in a dict alpha corrected by Sidak and Bonferroni to ease concat
alpha_SidakBonf = dict(alphacSidak=FDR_corr[2], alphacBonf=FDR_corr[3])
# select two first elements, transpose and rename columns
FDR_corr = pd.DataFrame(
FDR_corr[0:2]).transpose().rename(columns={
0: 'FDR',
1: 'pvals_corrected'
})
# add columns of p-values with FDR correction to df
df_corr_FDR = pd.concat([df_corr, FDR_corr],
axis=1,
join='outer',
ignore_index=False)
# add cloumns with corrected p-values
df_corr_FDR['alphacSidak'], df_corr_FDR['alphacBonf'] = alpha_SidakBonf[
'alphacSidak'], alpha_SidakBonf['alphacBonf']
# list of dfs, each df contains correlations for each frequency band
list_of_dfs.append(df_corr_FDR)
# concat list of dfs into 1 df
# reset index. After concat remains the index of each df on its own rows
df = pd.concat(list_of_dfs).reset_index(drop=True)
return df
#Do people who use Signal use security in choosing
#instant messaging tools?
#Simple row-to-row correlation
pg.corr(x=df['Q3-17'], y=df['Q34-31'])
# In[ ]:
pg.corr(x=df['Q40-0'], y=df['Q3-16'])
# In[ ]:
corr = pg.pairwise_corr(df,
columns=[['Q7-7'],
[
'Q3-0', 'Q3-1', 'Q3-2', 'Q3-3', 'Q3-4',
'Q3-5', 'Q3-6', 'Q3-7', 'Q3-8', 'Q3-9',
'Q3-10', 'Q3-11', 'Q3-12', 'Q3-13',
'Q3-14', 'Q3-15', 'Q3-16', 'Q3-17',
'Q3-18'
]],
method='pearson')
corr.sort_values(by=['p-unc'])[['X', 'Y', 'n', 'r', 'p-unc']].head()
# shows that Signal, IMO and Telegram are associated with users who choose baesd on security
# In[ ]:
# It appears largest correlations between tool usage and why it was selected are for security, features and work/school
corr = pg.pairwise_corr(df,
columns=[[
'Q7-0', 'Q7-1', 'Q7-2', 'Q7-3', 'Q7-4', 'Q7-5',
'Q7-6', 'Q7-7', 'Q7-8', 'Q7-9', 'Q7-10'
],
#Plot histogram of data: data.hist(bins=50, figsize=(18,14)) #Plot scatter matrix of data: scatter_matrix(data, figsize=(18,14)) #Find correlation matrix of data: corr_matrix = data.corr() #Print out correlations of Output w.r.t. others: print(corr_matrix[data.columns[0]].sort_values(ascending=False)) # In[20]: import pingouin as pg pg.pairwise_corr(data).sort_values(by=['p-unc'])[['X', 'Y', 'n', 'r', 'p-unc']].head() # In[21]: #Pairwise correlation import pingouin as pg from pingouin import pairwise_corr, read_dataset data[['Output (X0)', 'Tourism Score (X2)', 'Economic Score (X1)','SARS Score (X3)', 'Human Freedom Score']].pcorr() # In[22]: # Partial Correlations Matrix of variables
