def anova_analysis():
data_dropped_na = data.dropna()
mc1 = multi.MultiComparison(data_dropped_na[var_formula.get().split('~')[0]], data_dropped_na[var_formula.get().split('~')[1]])
result = mc1.tukeyhsd()
t = result.summary().as_text()
a_list = t.split('\n')
cols = [col for col in a_list[2].split(' ') if col]
df = pd.DataFrame(columns=cols)
for i in range(4, len(a_list) - 1):
items = [item for item in a_list[i].split(' ') if item]
df.loc[i-4] = items
formula = var_formula.get()
mod = ols(formula, data=data_dropped_na).fit()
# print(mod.summary())
aov_table = sm.stats.anova_lm(mod, typ=2)
writer = pd.ExcelWriter('Analysis/ANOVA.xlsx')
caption = pd.DataFrame(columns=[var_formula.get().split('~')[0]])
caption.to_excel(writer, sheet_name='Sheet1')
aov_table.to_excel(writer, sheet_name='Sheet1', startcol=1)
df.to_excel(writer, sheet_name='Sheet1', startcol=7)
writer.save()
os.startfile('Analysis\ANOVA.xlsx')
def test_using_anova(model,
model_results,
homoskedastic,
df,
dependent_var,
*independent_vars,
anova_type=2):
"""
Generate and ANOVA table and test the results using that
"""
output = ''
results = {}
aov_table = sm.stats.anova_lm(model_results,
typ=anova_type,
robust=None if homoskedastic else 'hc3')
aov_table = augment_anova_table(aov_table)
output += f"ANOVA\n{aov_table}\n\n"
results['anova'] = aov_table
# Perform multiple comparisons
mc = multicomp.MultiComparison(
df[dependent_var], df.loc[:,
independent_vars].astype(str).agg(','.join,
axis=1))
mc_results = mc.tukeyhsd()
output += f"Tukey's HSD:\n{mc_results}\n\n"
results['multiple'] = mc_results
return output, results
def bonferroni(item):
modelData = data[['Condition', item]].copy()
modelData['Condition'] = modelData['Condition'].map(str)
comp = mc.MultiComparison(modelData[item], modelData['Condition'])
tbl, a1, a2 = comp.allpairtest(stats.ttest_ind, method="bonf")
print(tbl)
def tukeys_test(self, Features=None, Clstrs=None, alpha=0.05):
"""
Tukey’s range test to compare means of all pairs of groups
Parameters
----------
Features: 2D_array_like
The arrays must have the same shape, except in the dimension
Clstrs: array_like
alpha: Value of FWER at which to calculate HSD
Returns
-------
A results class containing relevant data and some post-hoc calculations
"""
if Features is None:
Features = self.__data.columns[:-1].copy()
if Clstrs is None:
Clstrs = self.__data["Clusters"].copy()
Clstrs = Clstrs.dropna().unique().tolist()
Clstrs.sort()
for feature in Features:
print("\n\n", feature,"\n")
sub = self.__data[[feature, "Clusters"]].copy()
sub = sub.dropna()
mc = multi.MultiComparison(sub[sub["Clusters"].isin(Clstrs)][feature],
sub[sub["Clusters"].isin(Clstrs)]["Clusters"])
res = mc.tukeyhsd(alpha)
print(res.summary(),"\n\n")
def tukey(item):
modelData = data[['Condition', item]].copy()
modelData['Condition'] = modelData['condition'].map(str)
comp = mc.MultiComparison(modelData[item], modelData['condition'])
post_hoc_res = comp.tukeyhsd()
post_hoc_res.summary()
print(post_hoc_res.summary())
def compare_many(data):
'''Multiple comparisons: Which one is different? '''
print('\n MultComp: --------------------------------------')
# An ANOVA is a hypothesis test, and only answers the question: "Are all the groups
# from the same distribution?" It does not tell you which one is different.
# Since we now compare many different groups to each other, we have to adjust the
# p-values to make sure that we don't get a Type I error. For this, we use the
# statscom module "multicomp"
mc = multicomp.MultiComparison(data['weight'], data['group'])
# There are many ways to do multiple comparisons. Here, we choose
# "Tukeys Honest Significant Difference" test
# The first element of the output ("0") is a table containing the results
print(mc.tukeyhsd().summary())
# Show the group names
print(mc.groupsunique)
# Generate a print ----------------
res2 = mc.tukeyhsd() # Get the data
simple = False
if simple:
# You can do the plot with a one-liner, but then this does not - yet - look that great
res2.plot_simultaneous()
else:
# Or you can do it the hard way, i.e. by hand:
# Plot values and errorbars
xvals = np.arange(3)
plt.plot(xvals, res2.meandiffs, 'o')
errors = np.ravel(np.diff(res2.confint)/2)
plt.errorbar(xvals, res2.meandiffs, yerr=errors, fmt='o')
# Set the x-limits
xlim = -0.5, 2.5
# The "*xlim" passes the elements of the variable "xlim" elementwise into the function "hlines"
plt.hlines(0, *xlim)
plt.xlim(*xlim)
# Plot labels (this is a bit tricky):
# First, "np.array(mc.groupsunique)" makes an array with the names of the groups;
# and then, "np.column_stack(res2[1][0])]" puts the correct groups together
pair_labels = mc.groupsunique[np.column_stack(res2._multicomp.pairindices)]
plt.xticks(xvals, pair_labels)
plt.title('Multiple Comparison of Means - Tukey HSD, FWER=0.05' +
'\n Pairwise Mean Differences')
plt.show()
def anova_analysis():
if var_formula.get() == '':
print_status("Warning: Formula is missing", 'red')
return
data_dropped_na = data.dropna()
continuous_columns = [
col for col in data_dropped_na.columns
if data_dropped_na[col].dtype != 'object'
]
col = 1
writer = pd.ExcelWriter('../../Analysis/ANOVA.xlsx')
if '.' in var_formula.get():
dependent_vars = continuous_columns
else:
dependent_vars = var_formula.get().split('~')[0].split('+')
for dependent in dependent_vars:
mc1 = multi.MultiComparison(
data_dropped_na[dependent],
data_dropped_na[var_formula.get().split('~')[1]])
result = mc1.tukeyhsd()
t = result.summary().as_text()
a_list = t.split('\n')
cols = [col for col in a_list[2].split(' ') if col]
df = pd.DataFrame(columns=cols)
for i in range(4, len(a_list) - 1):
items = [item for item in a_list[i].split(' ') if item]
df.loc[i - 4] = items
formula = dependent + '~' + var_formula.get().split('~')[1]
mod = ols(formula, data=data_dropped_na).fit()
aov_table = sm.stats.anova_lm(mod, typ=2)
caption = pd.DataFrame(columns=[dependent])
caption.to_excel(writer, sheet_name='Sheet1', startrow=2, startcol=col)
aov_table.to_excel(writer,
sheet_name='Sheet1',
startcol=col,
startrow=3)
df.to_excel(writer, sheet_name='Sheet1', startrow=3, startcol=col + 7)
col += 16
writer.save()
os.startfile('../../Analysis\ANOVA.xlsx')
print_status('Status: Successful analysis', 'black')
def Stat_Test(df_gene, df_sg):
comp_groups = raw_input(
'Choose the groups you want to compare. (e.g. good/bad or good/no ...)'
)
tests = raw_input(
'Which statistical test you want to do? (e.g. unpaired_t_test or mann_whitney_test or fisher_exact_test or mann_whitney_test&fisher_exact_test ...)'
)
group_list = comp_groups.split('/')
test_list = tests.split('&')
if len(group_list) == 2:
group1 = group_list[0]
group2 = group_list[1]
list_g1 = list(df_sg.loc[df_sg.Group == group1, 'Sample'])
list_g2 = list(df_sg.loc[df_sg.Group == group2, 'Sample'])
df_gene = df_gene.loc[:, ['Gene'] + list_g1 + list_g2]
df_gene[group1 + '_Sum'] = df_gene.loc[:, list_g1].sum(axis=1)
df_gene[group2 + '_Sum'] = df_gene.loc[:, list_g2].sum(axis=1)
df_gene = df_gene.loc[(df_gene[group1 + '_Sum'] != 0) &
(df_gene[group2 + '_Sum'] != 0), :]
for t in test_list:
if t == 'unpaired_t_test':
df_gene['unpaired_t_test'] = df_gene.apply(
lambda row: unpaired_t_test(row, list_g1, list_g2), axis=1)
elif t == 'mann_whitney_test':
df_gene['mann_whitney_test'] = df_gene.apply(
lambda row: mann_whitney_test(row, list_g1, list_g2),
axis=1)
elif t == 'fisher_exact_test':
df_gene['fisher_exact_test'] = df_gene.apply(
lambda row: fisher_exact_test(row, list_g1, list_g2),
axis=1)
else:
print 'Please choose from unpaired_t_test, mann_whitney_test and fisher_exact_test'
sys.exit(1)
p_cor_flag = raw_input(
'Do you want to perform mutiple testing correction? (Y|N)')
if p_cor_flag == 'Y':
p_cor = raw_input(
'Which correction method do you want to choose? (e.g. bonferroni or fdr_bh or bonferroni&fdr_bh)'
)
p_cor = p_cor.split('&')
p_val = raw_input(
'Which p value do you want to correct? (e.g. unpaired_t_test or mann_whitney_test or fisher_exact_test'
)
for cor in p_cor:
df_gene[cor] = ssm.MultiComparison(df_gene[p_val],
alpha=0.05,
method=cor)
return df_gene
def tukey(data=None, independent=None, dependent=None):
pd.set_eng_float_format(accuracy=3, use_eng_prefix=False)
independent = str(independent)
dependent = str(dependent)
if input_check_numerical_categorical(data, independent, dependent):
return
test = multi.MultiComparison(data[dependent], data[independent])
res = test.tukeyhsd()
display(res.summary())
res.plot_simultaneous()
return
def anova(wine_set):
prepared_data = add_categ_quality(wine_set)
model1 = smf.ols(formula="total_sulfur_dioxide ~ C(quality_mark)", data=prepared_data)
results1 = model1.fit()
print(results1.summary())
#
sub = prepared_data[['total_sulfur_dioxide', 'quality_mark']]
print("\nMeans for total sulfur dioxide by quality marks of wine")
print(sub.groupby('quality_mark').mean())
print("\nStandard deviations for total sulfur dioxide by quality marks of wine")
print(sub.groupby('quality_mark').std(), '\n')
#
# # ------------- to perform Post hoc test (using Tukey's Honestly Significant Difference Test) -------------------------
mc1 = multi.MultiComparison(sub['total_sulfur_dioxide'], sub['quality_mark'])
res1 = mc1.tukeyhsd()
print(res1.summary())
def anova(wine_set):
prepared_data = add_categ_quality(wine_set)
model1 = smf.ols(formula="total_sulfur_dioxide ~ C(quality_mark)",
data=prepared_data)
results1 = model1.fit()
print(results1.summary())
#
sub = prepared_data[['total_sulfur_dioxide', 'quality_mark']]
print("\nMeans for total sulfur dioxide by quality marks of wine")
print(sub.groupby('quality_mark').mean())
print(
"\nStandard deviations for total sulfur dioxide by quality marks of wine"
)
print(sub.groupby('quality_mark').std(), '\n')
mc1 = multi.MultiComparison(sub['total_sulfur_dioxide'],
sub['quality_mark'])
res1 = mc1.tukeyhsd()
print(res1.summary())
def test_using_kruskal(df,
dependent_var,
*independent_vars,
correction_method='bonf'):
"""
Test for the significance of factors using the non-parametric Kruskal-Wallis
test followed by a Mann-Whitney U test with Bonferroni correction
"""
output = ''
results = {}
unique_values = df.groupby(
list(independent_vars)).size().reset_index().rename(
columns={0: 'count'})
test_data = []
for row in unique_values.itertuples(index=False):
if len(independent_vars) > 1:
selectors = [(df[v] == getattr(row, v)) for v in independent_vars]
row_selector = np.logical_and(*selectors[:2])
for idx in range(2, len(independent_vars)):
row_selector = np.logical_and(row_selector, selectors[idx])
else:
v = independent_vars[0]
row_selector = df[v] == getattr(row, v)
test_data.append(df.loc[row_selector, dependent_var])
assert len(test_data) == unique_values.shape[0]
test_results = spstats.kruskal(*test_data)
output += f"Kruskal-Wallis test:\n{test_results.statistic}, {test_results.pvalue}\n\n"
results['kruskal'] = test_results
# Perform multiple comparisons with Bonferroni correction
try:
mc = multicomp.MultiComparison(
df[dependent_var],
df.loc[:, independent_vars].astype(str).agg(','.join, axis=1))
mc_results = mc.allpairtest(spstats.mannwhitneyu,
method=correction_method)
output += f"Pairwise Mann-Whitney U:\n{mc_results[0]}\n\n"
results['multiple'] = mc_results[0]
except Exception as e:
print("ERROR:", e)
return output, results
def anova_test(df):
col1 = 'job'
col2 = 'duration'
duration_frame = df[[col1, col2]].copy()
groups = duration_frame.groupby(col1)
#Job-groups
admin = groups.get_group('admin.')['duration']
bluecollar = groups.get_group('blue-collar')['duration']
student = groups.get_group('student')['duration']
housemaid = groups.get_group('housemaid')['duration']
services = groups.get_group('services')['duration']
unemployed = groups.get_group('unemployed')['duration']
entrepreneur = groups.get_group('entrepreneur')['duration']
selfemployed = groups.get_group('self-employed')['duration']
retired = groups.get_group('retired')['duration']
print('Admin:\n')
print(admin.head())
F, p = ss.f_oneway(admin, bluecollar, student, housemaid, services,
unemployed, entrepreneur, selfemployed, retired)
print('{} {}\n'.format(F, p))
if p < 0.05:
print(
'Null hypothesis rejected. Statistical difference found. Conduct post-hoc tests.'
)
else:
print('Not much difference found. Can accept null hypothesis.')
#Conducting a Post-Hoc test
duration_frame[col2] = duration_frame[col2].convert_objects(
convert_numeric=True)
mc = multi.MultiComparison(duration_frame[col2], duration_frame[col1])
result = mc.tukeyhsd()
print(result.summary())
def run_Tukey(self, t, name_list, z):
# Run tukey for data at t,
# the name_list is the naming of the columns
# z is the score to run Tukey
total_list = []
total_list2 = []
for i, j in zip(t, name_list):
total_list = np.concatenate((total_list, np.reshape(i, len(i))))
total_list2.extend([j] * len(i))
num_array = np.asarray(total_list)
symbol_group = np.asarray(total_list2)
di = {'Score': num_array, 'Group': symbol_group}
middle_frame = pd.DataFrame(di)
mcobj = ml.MultiComparison(
pd.to_numeric(middle_frame.Score, errors="ignore"),
middle_frame.Group)
out = mcobj.tukeyhsd(z)
return out
pokerhand_test=data_test['CLASS']
# put into a pandas dataFrame
pokerhand_train=pd.DataFrame(pokerhand_train)
pokerhand_test=pd.DataFrame(pokerhand_test)
pokerhand_train.reset_index(level=0, inplace=True) # reset index
merged_train_all=pd.merge(pokerhand_train, merged_train, on='index') # merge the pokerhand train with merged clusters
sub1 = merged_train_all[['CLASS', 'cluster']].dropna()
import statsmodels.formula.api as smf
import statsmodels.stats.multicomp as multi
# respone formula
pokermod = smf.ols(formula='CLASS ~ cluster', data=sub1).fit()
print (pokermod.summary())
print ('means for Poker hands by cluster')
m1= sub1.groupby('cluster').mean()
print (m1)
print ('standard deviations for Poker hands by cluster')
m2= sub1.groupby('cluster').std()
print (m2)
mc1 = multi.MultiComparison(sub1['CLASS'], sub1['cluster'])
res1 = mc1.tukeyhsd()
print(res1.summary())
# validate clusters in training data by examining cluster differences in GPA using ANOVA
# first have to merge GPA with clustering variables and cluster assignment data
gpa_data = data_clean['internetuserate']
# split GPA data into train and test sets
gpa_train, gpa_test = train_test_split(gpa_data,
test_size=.3,
random_state=123)
gpa_train1 = pd.DataFrame(gpa_train)
gpa_train1.reset_index(level=0, inplace=True)
merged_train_all = pd.merge(gpa_train1, merged_train, on='index')
sub1 = merged_train_all[['internetuserate', 'cluster']].dropna()
import statsmodels.formula.api as smf
import statsmodels.stats.multicomp as multi
gpamod = smf.ols(formula='internetuserate ~ (cluster)', data=sub1).fit()
print(gpamod.summary())
print('means for internetuserate by cluster')
m1 = sub1.groupby('cluster').mean()
print(m1)
print('standard deviations for internetuserate by cluster')
m2 = sub1.groupby('cluster').std()
print(m2)
mc1 = multi.MultiComparison(sub1['internetuserate'], sub1['cluster'])
res1 = mc1.tukeyhsd()
print(res1.summary())
cols = ['sum_sq', 'df', 'mean_sq', 'F', 'PR(>F)', 'eta_sq', 'omega_sq']
aov = aov[cols]
print(aov)
print(" ")
print(" POST-HOST TESTING ")
# POST-HOC TESTING
print(" ")
# TUKEY HONESTLY SIGNIFFICANCE DIFFERENCE
print("TUKEY HONESTLY SIGNIFFICANCE DIFFERENCE")
#from statsmodels.stats.multicomp import (pairwise_tukeyhsd,
# MultiComparison)
import statsmodels.stats.multicomp as mc
comp = mc.MultiComparison(df['Scores'], df['Method'])
post_hoc_res = comp.tukeyhsd()
post_hoc_res.summary()
print(post_hoc_res)
post_hoc_res.plot_simultaneous(ylabel="Method", xlabel="Score Differences")
#BONFERRONI CORRECTION
print(" ")
print("BONFERRONI CORRECTION")
print(" ")
import statsmodels.stats.multicomp as mc
comp = mc.MultiComparison(df['Scores'], df['Method'])
tbl, a1, a2 = comp.allpairtest(stats.ttest_ind, method="bonf")
print(tbl)