def chi_squared_test(self, p_columnDEP, p_columnINDEP, p_maxNumAttributes = 6):
tableREL, results, exp = rp.crosstab(p_columnDEP, p_columnINDEP, prop='col', test= 'chi-square', expected_freqs= True)
tableABS = rp.crosstab(p_columnDEP, p_columnINDEP)
ct = pd.crosstab(p_columnDEP, p_columnINDEP, normalize ='columns').reset_index()
ct = ct.transpose()
stacked = ct.stack().reset_index().rename(columns={0:'value'})
stacked = eval('stacked[stacked.' + p_columnINDEP.name + '!= p_columnDEP.name]')
stacked = stacked.rename(columns={"level_1": p_columnDEP.name})
_myPlot = eval('sns.barplot(x=stacked.' + p_columnINDEP.name + ', y=stacked.value, hue=stacked.' + p_columnDEP.name + ')')
a = str(len(tableABS.columns) - 1) + ' attributes'
a0 = '--------------------------------------------------------------------------------------'
a1 = '----------------------------------------------------------------- OBSERVED FREQUENCIES'
a2 = '------------------------------------------------------ OBSERVED FREQUENCIES (relative)'
a3 = '----------------------------------------------------------------- EXPECTED FREQUENCIES'
a4 = '------------------------------------------------------------------------- TEST RESULTS'
a5 = '--------------------------------------------------------------------------------- PLOT'
if len(tableABS.columns) - 1 > p_maxNumAttributes:
b1 = '--- not plotted --- number of attributes with ' + str(len(tableABS.columns) - 1) + ' is too big ---------------------------'
return [a, a0, a1, a0, b1, a2, b1, a3, b1, a4, results, a5, _myPlot]
else:
return [a, a0, a1, a0, tableABS, a0, a2, a0, tableREL, a0, a3, a0, round(exp,0), a0, a4, a0, results, a0, a5, a0, _myPlot]
def RQ1(vul_dict):
frame = {'fix_landing': [], 'client_landing': []}
counter = {'major': 0, 'minor': 0, 'patch': 0}
for vul in vul_dict:
fix_landing = vul_dict[vul]['package']['release_type']
counter[fix_landing] += 1
for client in vul_dict[vul]['client']:
if client['adoption'] != None:
frame['fix_landing'].append(fix_landing + '_landing')
frame['client_landing'].append(
client['adoption'] + '_client_landing'
if client['adoption'] != 'removal' else 'removal')
df = pd.DataFrame(frame)
crosstab, res = researchpy.crosstab(df['client_landing'],
df['fix_landing'],
test='chi-square')
print('=' * 50)
print(
'(RQ1) A contingency table shows the frequency distribution of client-side fixing release update for each fixing release update.'
)
print('=' * 50)
print(crosstab)
print(res)
print(counter)
print('=' * 50)
print()
def do_different_samples_match(test):
core = ThesisCore(in_package=True)
core.create_reprs(kmer=3)
for name, repr in core._repr_core.reprs.items():
temp_df = repr.repr_df.copy()
for indi, i in enumerate(range(0, 6000, 600)):
for indj, j in enumerate(range(0, 6000, 600)):
if indj <= indi:
continue
temp_df1, temp_df2 = temp_df[i:i + 600], temp_df[j:j + 600]
if test == 'permutation':
p_value = get_permutation_of_pdists(temp_df1, temp_df2)
print(p_value)
elif test == 'KS':
p_value = get_KS_of_pdists(temp_df1, temp_df2)
if p_value > 0.05:
print('we Cannot reject the H0 that the 2 dists', indi,
indj, 'are identical with pvalue', p_value)
else:
print('we Reject the H0 that the 2 dists ', indi, indj,
'are identical with pvalue', p_value)
elif test == 'crosstab':
res = researchpy.crosstab(temp_df1,
temp_df2,
test="chi-square")
print(res)
print(res['p-value'])
def _get_prop_vs_prop_dss_score(treatment, con, dfs=None, use_percentage=None, use_labels=None, metric=None, as_percentage=None, is_categorical=None, alternative=None):
df_prop, df_resp = get_prop_resp(treatment)
df_prop_ref, df_resp_ref = get_prop_resp("t12a")
prop_values = metric(df_resp["min_offer"], df_prop["offer"])
prop_value = metrics.get_mean(prop_values)
prop_dss_values = metric(df_resp["min_offer"], df_prop["offer_dss"])
prop_dss_value = metrics.get_mean(prop_dss_values)
prop_values_ref = metric(df_resp_ref["min_offer"], df_prop_ref["offer_dss"])
prop_value_ref = metrics.get_mean(prop_values_ref)
# auto_dss_values = metric(df_resp["min_offer"], df_prop["ai_offer"])
# auto_dss_value = metrics.get_mean(auto_dss_values)
dof = 0
diff = None
if is_categorical:
table, res = rp.crosstab(pd.Series(prop_values_ref), pd.Series(prop_dss_values), test='chi-square')
s, p, r = res.results.values
test_label = f"(pearson chi2)"
test_label = f"chi2"
print("Conclusion: ", generate_cat_stat_sentence(np.mean(resp_dss_values), np.std(resp_dss_values), np.mean(auto_dss_values), np.std(auto_dss_values), s, p, dof, diff=diff, label1=treatment+".dss", label2="t20.dss"))
else:
table, res = rp.ttest(pd.Series(prop_values_ref), pd.Series(prop_dss_values), paired=False)
s = res.results[2]
if alternative=="greater":
p = res.results[4]
elif alternative == "less":
p = res.results[5]
elif alternative in (None, 'two-sided'):
p = res.results[3]
r = res.results[9]
diff = res.results[0]
dof = res.results[1]
print("Conclusion: ", generate_stat_sentence(np.mean(resp_dss_values), np.std(resp_dss_values), np.mean(auto_dss_values), np.std(auto_dss_values), s, p, dof, diff=diff, label1=treatment+".dss", label2="t20.dss"))
test_label = f"(ttest independent) H0: {'equal' if alternative in {None, 'two-sided'} else alternative}"
print("RESUME: ", res)
if as_percentage:
res = {
"Proposer + DSS": f'{100 * prop_dss_value:.2f} %',
"T10": f'{100 * prop_value_ref:.2f} %',
}
else:
res = {
"Proposer + DSS": f'{prop_dss_value:.2f}',
"T10": f'{prop_value_ref:.2f}',
}
if is_categorical:
res[test_label] = f"{s:.3f} (p: {p:.3f}, phi: {r:.3f})"
else:
res[test_label] = f"{s:.3f} (p: {p:.3f}, r: {r:.3f})"
return res
def cramersV_matrix(catFeat, X):
'''
Function that takes in the list of categorical features and returns
the Cramer's V matrix which is similar to chi^2 except it adjusts for
sample size
'''
assocMatrix = []
for col1 in catFeat:
for col2 in catFeat:
tab, res = rp.crosstab(X[col1], X[col2], test = 'chi-square')
assocMatrix.append(res.iloc[2,1])
assocMatrix = np.array(assocMatrix).reshape((len(catFeat),-1))
return assocMatrix
def chi_square_test_for_each_category_level(self, p_columnDEP, p_columnINDEP):
g = pd.get_dummies(p_columnINDEP)
h = pd.DataFrame(p_columnDEP)
k = h.join(g)
k = k.reset_index().drop(['index'], axis=1)
temp = pd.DataFrame(columns = ['AttrCharacteristic','AttrCharacteristicRenamed','ChiSq','p_value','CramerV'])
for i in range(1,len(k.columns)):
_from = k.columns[i]
_to = re.sub('[^A-Za-z0-9_]+', '_', k.columns[i])
k = k.rename(columns = { _from: _to })
a, results = rp.crosstab(k[p_columnDEP.name], k[k.columns[i]], test= 'chi-square')
temp = temp.append({'AttrCharacteristic': _from,
'AttrCharacteristicRenamed': _to,
'ChiSq': results.iloc[0,1],
'p_value': results.iloc[1,1],
'CramerV': results.iloc[2,1]}
, ignore_index=True)
return temp.sort_values(by='CramerV', ascending=False, na_position='first')
def apply_chi_squared_statistic_on_timeline(self, p_df, p_columnDEP, p_columnINDEP, p_dates):
temp = p_df.copy(deep=True)
temp2 = None
temp_i = None
function_start = time.time()
print('Starting:' + '\t' + 'date: ' + '\t' + time.ctime())
# create empty dataset
temp2 = pd.DataFrame(columns = ['Starting_Date', 'ChiSq', 'p_value', 'CramerV'])
for i in p_dates:
starting = time.time()
temp_i = temp[temp['starting_Date'] == i]
ct, results = rp.crosstab(temp_i[p_columnDEP], temp_i[p_columnINDEP], test= 'chi-square')
temp2 = temp2.append({ 'Starting_Date': i,
'ChiSq': results.iloc[0,1],
'p_value': results.iloc[1,1],
'CramerV': results.iloc[2,1]}, ignore_index=True)
print('\n' + 'Duration: ' + '\t' + '{:5.3f}s'.format(time.time() - function_start) + '\n')
return temp2
import pandas as pd
import researchpy as rp
import matplotlib.pyplot as plt
df = pd.read_csv('imdb_data_clean.csv', delimiter=';')
# CROSSTABULATION
table, results = rp.crosstab(df['sfxgenre'],
df['productionlocation'],
prop='col',
test='chi-square')
print(table) # Prints crosstab
print()
print(results) # Prints statistics tab
# STACKED BAR CHART
ct = pd.crosstab(df['productionlocation'], df['sfxgenre'],
normalize='index') # Create PD crosstab as basis for plot
ct.plot.bar(stacked=True) # Generate plot
plt.legend(loc='lower left', bbox_to_anchor=(0.0, 1.01),
frameon=False) # Tidy up legend (other)
plt.xticks(rotation="horizontal") # Horizontal x-axis labels
plt.xlabel('Production location') # Adding label on x-axis
plt.ylabel('Proportion SFX') # Adding label on y-axis
plt.savefig('stackedbar.pdf') # Saving figure
plt.clf() # Clearing figure
import pandas as pd
import researchpy as rp
import matplotlib.pyplot as plt
from scipy import stats
df = pd.read_csv(
'http://www.digitalanalytics.id.au/static/files/youtube_vevo_clean.csv',
delimiter=',')
# Set max of rows to show, in/decrease to needs
pd.set_option('max_rows', 9999)
# CROSSTABULATION
table, results = rp.crosstab(df['view_cat'],
df['lyric_video'],
prop='col',
test='chi-square')
print(table) # Prints crosstab
print()
print(results) # Prints statistics tab
# STACKED BAR CHART
ct = pd.crosstab(df['lyric_video'], df['view_cat'],
normalize='index') # Create PD crosstab as basis for plot
ct.plot.bar(stacked=True) # Generate plot
plt.legend(loc='lower left', bbox_to_anchor=(0.0, 1.01),
frameon=False) # Tidy up legend (other)
plt.xticks(rotation="horizontal") # Horizontal x-axis labels
plt.xlabel('Lyric video') # Adding label on x-axis
plt.ylabel('Viewer categories') # Adding label on y-axis
# print(crosstab)
# print(res)
dfNormailTrail = df[df['noiseNumber'] != 'special']
dfSpecialTrail = df[df['noiseNumber'] == 'special']
statDF = pd.DataFrame()
statDF['firstIntentionConsistFinalGoalNormal'] = dfNormailTrail.groupby(
'name')["firstIntentionConsistFinalGoal"].mean()
statDF['firstIntentionConsistFinalGoalSpecail'] = dfSpecialTrail.groupby(
'name')["firstIntentionConsistFinalGoal"].mean()
# chi-squre
resultDf = dfSpecialTrail
crosstab, res = researchpy.crosstab(
resultDf['participantsType'],
resultDf['firstIntentionConsistFinalGoal'],
test="chi-square")
print(crosstab)
print(res)
# χ2 (2) = 14.93, P < 0.001, VCramer = 0.50
# fisher
# resultDf = resultDf[resultDf['participantsType'] != "machine"]
# crosstab, res = researchpy.crosstab(resultDf['participantsType'], resultDf['firstIntentionConsistFinalGoal'], test='fisher')
# print(crosstab)
# print(res)
#
df['trialType'] = [
'Critical Disruption' if trial == "special" else 'Random Disruptions'
participants = ['human', 'RL']
dataPaths = [os.path.join(resultsPath, participant) for participant in participants]
dfList = [pd.concat(map(pd.read_csv, glob.glob(os.path.join(dataPath, '*.csv'))), sort=False) for dataPath in dataPaths]
df = pd.concat(dfList)
df['participantsType'] = ['machine' if 'max' in name else 'human' for name in df['name']]
df['isDecisionStepInZone'] = df.apply(lambda x: isDecisionStepInZone(eval(x['trajectory']), eval(x['target1']), eval(x['target2']), x['decisionSteps']), axis=1)
dfExpTrail = df[(df['targetDiff'] == 0) & (df['conditionName'] == 'expCondition') & (df['decisionSteps'] != 10) & (df['decisionSteps'] != 6)]
dfExpTrail['hasAvoidPoint'] = dfExpTrail.apply(lambda x: isTrajHasAvoidPoints(eval(x['trajectory']), eval(x['aimAction']), eval(x['playerGrid']), eval(x['target1']), eval(x['target2']), x['decisionSteps'], x['conditionName']), axis=1)
resultDf = df[df['participantsType'] == 'human']
crosstab, res = researchpy.crosstab(resultDf['hasAvoidPoint'], resultDf['decisionSteps'], test="chi-square")
print(crosstab)
print(res)
# human vs machine in diff decision steps[2,4,6]
# decisionStep = 4
# dfExpTrail = df[(df['decisionSteps'] == decisionStep) & (df['targetDiff'] == 0) & (df['conditionName'] == 'expCondition')]
# dfExpTrail['hasAvoidPoint'] = dfExpTrail.apply(lambda x: isTrajHasAvoidPoints(eval(x['trajectory']), eval(x['aimAction']), eval(x['playerGrid']), eval(x['target1']), eval(x['target2']), x['decisionSteps'], x['conditionName']), axis=1)
resultDf = dfExpTrail
crosstab, res = researchpy.crosstab(resultDf['participantsType'], resultDf['isDecisionStepInZone'], test="chi-square")
print(crosstab)
print(res)
df["firstIntentionConsistFinalGoal"] = df.apply(lambda x: calculateFirstIntentionConsistency(eval(x['goal'])), axis=1)
# %%
#chi-square analysis
chi_exp = experiment_results.apply(share.count_frequencies, axis=1)[10:13]
chi_control = control_results.apply(share.count_frequencies, axis=1)[10:13]
chi_exp_ag = chi_exp.sum(axis=0)
chi_control_ag = chi_control.sum(axis=0)
chi = pd.concat([chi_exp_ag, chi_control_ag], axis=1, keys=['exp', 'con'])
chi2, p, dof, expected = chi2_contingency(chi)
# %%
#draw bar chart for one variable
data = chi_exp_ag.to_dict()
names = list(data.keys())
values = list(data.values())
fig, axs = plt.subplots(figsize=(5, 3))
axs.bar(names, values)
#%%c
#check the chi-sqaure between perception1 and perception2
crosstab, res, expected = researchpy.crosstab(
experiment_results.loc['perception1'],
experiment_results.loc['perception2'],
test='chi-square',
expected_freqs=True)
# %%
transformed_data = transformed_data.drop(['campaign'],axis = 1)
transformed_data['campaign'] = np.log(cleaned_data['campaign']).replace([np.inf, -np.inf], 0)
transformed_data['campaign'].skew()
for i in numeric_col:
print (i,':',transformed_data[i].skew())
# Bivariata analysis
## catagorical vaiable
import researchpy as rp
category=['job','education','marital','default','housing','loan','contact','month','poutcome']
for c in category :
table, results = rp.crosstab(data[c],data['y'], test= 'chi-square')
print(results)
print('='*45)
# croosstab with catagorical
for c in category :
table = pd.crosstab(data[c],data['y'])
table.plot(kind='bar')
# for numeric variable
corelation = data.corr()
ax=plt.subplots(figsize=(9,7))
sns.heatmap(corelation,annot = True)
### multivaite analysis
def _get_prop_vs_prop_dss_score(treatment,
con,
dfs=None,
use_percentage=None,
use_labels=None,
metric=None,
as_percentage=None,
is_categorical=None,
alternative=None):
df_prop, df_resp = get_prop_resp(treatment)
df_prop_t10, df_resp_t10 = get_prop_resp("t10a")
metric_values = metric(df_prop)
metric_value = metrics.get_mean(metric_values)
metric_t10_values = metric(df_prop_t10)
metric_value_t10 = metrics.get_mean(metric_t10_values)
metric_values = metrics.get_data(metric_values)
metric_t10_values = metrics.get_data(metric_t10_values)
#print(stats.chisquare(metric_values[:103], metric_t10_values[:103]))
dof = 0
diff = None
print(metric.__name__)
if is_categorical:
#table, res = rp.crosstab(pd.Series(metric_values), pd.Series(metric_t10_values), test='g-test')
table, res = rp.crosstab(pd.Series(metric_values),
pd.Series(metric_t10_values),
test='fisher')
#print(table, res)
#s, p, r = res.results
s = res.results[0]
p = res.results[1]
r = res.results[4]
test_label = f"(g-test chi2)"
print(
"Conclusion: ",
generate_cat_stat_sentence(np.mean(metric_t10_values),
np.std(metric_t10_values),
np.mean(metric_values),
np.std(metric_values),
s,
p,
dof,
diff=diff,
label1="t10a.dss",
label2=treatment + ".dss"))
print(
pd.crosstab(pd.Series(metric_t10_values),
pd.Series(metric_values)))
else:
table, res = rp.ttest(pd.Series(metric_t10_values),
pd.Series(metric_values),
paired=False)
s = res.results[2]
if alternative == "greater":
p = res.results[4]
elif alternative == "less":
p = res.results[5]
elif alternative in (None, 'two-sided'):
p = res.results[3]
r = res.results[9]
diff = res.results[0]
dof = res.results[1]
s = res.results[2]
p = res.results[3]
r = res.results[9]
print(
"Conclusion: ",
generate_stat_sentence(np.mean(metric_t10_values),
np.std(metric_t10_values),
np.mean(metric_values),
np.std(metric_values),
s,
p,
dof,
diff=diff,
label1="t10a.dss",
label2=treatment + ".dss"))
test_label = f"(ttest independent) H0: {'equal' if alternative in {None, 'two-sided'} else alternative}"
print("TABLE: ", table)
print("TEST: ", res)
if as_percentage:
res = {
"Proposer + DSS": f'{100 * metric_value:.2f} %',
"T10": f'{100 * metric_value_t10:.2f} %',
}
else:
res = {
"Proposer + DSS": f'{metric_value:.2f}',
"T10": f'{metric_value_t10:.2f}',
}
if is_categorical:
res[test_label] = f"{s:.3f} (p: {p:.3f}, phi: {r:.3f})"
else:
res[test_label] = f"{s:.3f} (p: {p:.3f}, r: {r:.3f})"
return res
ctype_type = pd.crosstab(index=train_cat_df["Cover_Type"],
columns=train_cat_df["Wilderness_Area1"])
ctype_type
# In[26]:
stats.chi2_contingency(ctype_type)
# In[27]:
table, results = rp.crosstab(train_cat_df["Cover_Type"], train_cat_df["Wilderness_Area1"], prop= 'col', test= 'chi-square')
table
# In[28]:
print(results)
# In[29]:
for col in train_cat_df.columns[:-1]:
table, results = rp.crosstab(train_cat_df["Cover_Type"], train_cat_df[col], prop= 'col', test= 'chi-square')
if(results.results[1]<0.05):
def cramer(df):
table, results = rp.crosstab(df['image'],
df['type'],
prop='col',
test='chi-square')
else:
return ("N")
df_train["Age_above65_female"] = df_train[["Age", "Sex"]].apply(age_female,
axis=1)
print("變數_Ageabove65_female\n", df_train)
# Case03_透過昨天課程的內容，驗證產生的兩個新變數，哪一個和目標變數(Survived_cate) 的相關性較高？
# Survived_cate 和 兩變數(Age_above65、Age_above65_female)關係都是離散vs離散,因此採用Cramer’s V 係數
le = preprocessing.LabelEncoder()
df_train['Age_above65'] = le.fit_transform(df_train['Age_above65'])
above65_table = pd.crosstab(df_train["Age_above65"], df_train['Survived_cate'])
df_above65_table = min(above65_table.shape[0], above65_table.shape[1]) - 1
above65_table, res_above65 = researchpy.crosstab(df_train["Age_above65"],
df_train['Survived_cate'],
test='chi-square')
df_train['Age_above65_female'] = le.fit_transform(
df_train['Age_above65_female'])
above65_female_table = pd.crosstab(df_train["Age_above65_female"],
df_train['Survived_cate'])
df_above65_female_table = min(above65_female_table.shape[0],
above65_female_table.shape[1]) - 1
above65_female_table, res_above65_female = researchpy.crosstab(
df_train["Age_above65_female"],
df_train['Survived_cate'],
test='chi-square')
res = {}
res["Survived_vs_Age_above65"] = [
from itertools import permutations
import researchpy
cols = df.columns[:19]
perm = permutations(cols, 2)
outP = {"indeX": list(), "colS": list(), "valueS": list()}
ii = 0
for hh in list(perm)[67:]:
col1 = mm[hh[0]]
col2 = mm[hh[1]]
if mm.dtypes[hh[0]].name == "float64":
col1 = pd.qcut(col1, 5)
if mm.dtypes[hh[1]].name == "float64":
col2 = pd.qcut(col2, 5)
a, b = researchpy.crosstab(col1, col2, test="chi-square")
outP["valueS"].append(b.iloc[2, 1])
outP["indeX"].append(hh[0])
outP["colS"].append(hh[1])
ii += 1
print(ii)
rouP = pd.DataFrame(outP)
mmK = rouP.pivot_table(index="indeX", columns="colS", values="valueS")
mmK.fillna(1, inplace=True)
mmK.to_pickle("heatmap.pkl")
fig = plt.figure(figsize=(10, 10))
sns.heatmap(mmK, cmap="BuPu")
plt.savefig("heatmap_1.png")
def _get_prop_vs_prop_dss_score(treatment, con, dfs=None, use_percentage=None, use_labels=None, metric=None, as_percentage=None, is_categorical=None, alternative=None):
df_prop, df_resp = get_prop_resp(treatment)
prop_values = metric(df_resp["min_offer"], df_prop["offer"])
prop_value = metrics.get_mean(prop_values)
prop_dss_values = metric(df_resp["min_offer"], df_prop["offer_dss"])
prop_dss_value = metrics.get_mean(prop_dss_values)
auto_dss_values = metric(df_resp["min_offer"], df_prop["ai_offer"])
auto_dss_value = metrics.get_mean(auto_dss_values)
dof = 0
diff = None
print(metric)
if is_categorical:
table = pd.crosstab(prop_values, prop_dss_values)
# print("TABLE: ", table)
# checked using: http://vassarstats.net/propcorr.html
# s, p = sms2.mcnemar(prop_values, prop_dss_values, exact=False, correction=False)
table, res = rp.crosstab(prop_values, prop_dss_values, test='mcnemar')
#chi, p, s = (res.results.values)
s, p, r = (res.results.values)
print("Conclusion: ", generate_stat_sentence(np.mean(prop_values), np.std(prop_values), np.mean(prop_dss_values), np.std(prop_dss_values), s, p, dof, diff=diff, label1=treatment, label2=treatment+".dss"))
test_label = f"(mcnemar - chi2)"
else:
s, p = stats.wilcoxon(prop_values, prop_dss_values, alternative=alternative or 'two-sided')
table, res = rp.ttest(pd.Series(prop_values), pd.Series(prop_dss_values), paired=True)
#res = rp.ttest(pd.Series(prop_values), pd.Series(prop_dss_values), paired=True)
diff = res.results[0]
dof = res.results[1]
s = res.results[2]
p = res.results[3]
r = res.results[9]
test_label = f"(ttest dependent)"
print("Conclusion: ", generate_stat_sentence(np.mean(prop_values), np.std(prop_values), np.mean(prop_dss_values), np.std(prop_dss_values), s, p, dof, diff=diff, label1=treatment, label2=treatment+".dss"))
print("TABLE:", table)
print("RES:", res)
if as_percentage:
res = {
"Proposer": f'{100 * prop_value:.2f} %',
"Proposer + DSS": f'{100 * prop_dss_value:.2f} %',
# "prop:dss - prop": f'{100 * (prop_dss_value - prop_value):.2f} %',
}
else:
res = {
"Proposer": f'{prop_value:.2f}',
"Proposer + DSS": f'{prop_dss_value:.2f}',
# "prop:dss - prop": f'{(prop_dss_value - prop_value):.2f} %',
}
if is_categorical:
res[test_label] = f"{s:.3f} (p: {p:.3f}, phi: {r:.3f})"
else:
res[test_label] = f"{s:.3f} (p: {p:.3f}, r: {r:.3f})"
return res
def mcnemar(self, df):
"""
This function performs the mcnemar test and returns a df with the results
"""
table, res = rp.crosstab(df['perm'], df['t_test'], test='mcnemar')
return res
dv = "Age"
between = "Survived_cate"
aov = pg.anova(dv=dv, between=between, data=df_train, detailed=True)
etaSq = aov.SS[0] / (aov.SS[0] + aov.SS[1])
print("Q1: 透過數值法計算 Age 和 Survived 是否有相關性? A:連續與離散")
print("Eta Squared (η2)結果：%.3f 相關性 %s" % (etaSq, valiate_etaSq(etaSq)))
# print("Cramer's Values 結果 ", res.loc[2, 'results'], valiate_etaSq(res.loc[2, 'results']))
dt = 'Survived_cate'
between = 'Sex'
# step1: 用交叉列連表(contingency table)，來整理兩個類別型的資料
contTable = pd.crosstab(df_train[between], df_train[dt])
df = min(contTable.shape[0], contTable.shape[1]) - 1
crosstab, res = researchpy.crosstab(df_train[between],
df_train[dt],
test='chi-square')
print()
print("Q2:透過數值法計算 Sex 和 Survived 是否有相關性? A:離散與離散")
print("Cramer's 相關性%.3f 結果 %s" %
(res.loc[2, 'results'], judgment_CramerV(df, res.loc[2, 'results'])))
print()
# 連續與連續 Pearson
def judgment_PearsonV(corr):
if corr < .1:
qual = "無線性相關"
elif corr < .4:
qual = "低度線性相關"
elif corr < .7:
McNemar's Test
"""
xls = pd.ExcelFile("formcnemar18.xlsx")
sports = xls.sheet_names
df = {sport: xls.parse(sport) for sport in sports}
rankings = df["NHL"].columns.tolist()[-15:-1]
# mcnemar by league
mcnemar_league = {}
for sport in sports:
mcnemar_league[sport] = np.ones([len(rankings), len(rankings), 2])
for i, r1 in enumerate(rankings):
for j, r2 in enumerate(rankings):
if r1 != r2:
_, res = rp.crosstab(df[sport][r1],
df[sport][r2],
test="mcnemar")
chisq, p, _ = res.results.values
mcnemar_league[sport][i, j] = np.array([chisq, p])
# write to excel
writer = pd.ExcelWriter('mcnemar_league.xlsx', engine='xlsxwriter')
# define sheet names
chisq_sheets = [sport + "_chisq" for sport in sports]
p_sheets = [sport + "_p-value" for sport in sports]
# loop for each sport
for sport, chisq_sheet, p_sheet in zip(sports, chisq_sheets, p_sheets):
# chi_sq
chisq_df = pd.DataFrame(mcnemar_league[sport][:, :, 0],
index=rankings,
def _get_prop_vs_prop_dss_score(treatment,
con,
dfs=None,
use_percentage=None,
use_labels=None,
metric=None,
as_percentage=None,
is_categorical=None,
alternative=None):
df_prop, df_resp = get_prop_resp(treatment)
df_prop_t20, df_resp_t20 = get_prop_resp("t20a")
# prop_values = metric(df_resp["min_offer_dss"], df_prop["offer"])
# prop_value = metrics.get_mean(prop_values)
prop_dss_values = metric(df_resp["min_offer_dss"], df_prop["offer_dss"])
prop_dss_value = metrics.get_mean(prop_dss_values)
auto_dss_values = metric(df_resp_t20["min_offer_dss"],
df_prop_t20["ai_offer"])
auto_dss_value = metrics.get_mean(auto_dss_values)
dof = 0
diff = None
if is_categorical:
# table = np.array([np.bincount(prop_values), np.bincount(prop_dss_values)])
# print("TABLE: ", table)
# checked using: http://vassarstats.net/propcorr.html
# s, p = sms2.mcnemar(prop_values, prop_dss_values, exact=False, correction=False)
table, res = rp.crosstab(prop_dss_values,
auto_dss_values,
test='mcnemar')
s, p, r = res.results.values
test_label = f"(mcnemar) H0: equal, Ha: {'two-sided'}"
print(
"Conclusion: ",
generate_cat_stat_sentence(np.mean(prop_dss_values),
np.std(prop_dss_values),
np.mean(auto_dss_values),
np.std(auto_dss_values),
s,
p,
dof,
diff=diff,
label1=treatment + ".dss",
label2="t20.dss"))
else:
#s, p = stats.wilcoxon(prop_values, auto_dss_values, alternative=alternative or 'two-sided')
table, res = rp.ttest(pd.Series(prop_dss_values),
pd.Series(auto_dss_values),
paired=False)
test_label = f"(wilcoxon) H0: equal, Ha: {alternative or 'two-sided'}"
diff = res.results[0]
dof = res.results[1]
s = res.results[2]
p = res.results[3]
r = res.results[9]
print(
"Conclusion: ",
generate_stat_sentence(np.mean(prop_dss_values),
np.std(prop_dss_values),
np.mean(auto_dss_values),
np.std(auto_dss_values),
s,
p,
dof,
diff=diff,
label1=treatment + ".dss",
label2="t20.dss"))
if as_percentage:
res = {
# "Proposer": f'{100 * prop_value:.2f} %',
"Proposer + DSS":
f'{100 * prop_dss_value:.2f} %',
"T20 Auto DSS":
f'{100 * auto_dss_value:.2f} %',
"prop:dss - auto prop":
f'{100 * (prop_dss_value - auto_dss_value):.2f} %',
}
else:
res = {
# "Proposer": f'{prop_value:.2f}',
"Proposer + DSS": f'{prop_dss_value:.2f}',
"T20 Auto DSS": f'{auto_dss_value:.2f}',
"prop:dss - auto prop":
f'{(prop_dss_value - auto_dss_value):.2f} %',
}
if is_categorical:
res[test_label] = f"{s:.3f} (p: {p:.3f}, phi: {r:.3f})"
else:
res[test_label] = f"{s:.3f} (p: {p:.3f}, r: {r:.3f})"
return res
def get_info_accuracy(treatment, con, dfs=None, use_percentage=None, use_labels=None):
if treatment in ("t13a", "t13"):
ref = "t12a"
elif treatment in ("t11a", "t11b"):
ref = "t10b"
else:
ref = treatment
df_prop, df_resp = get_prop_resp(treatment)
df_prop_ref, df_resp_ref = get_prop_resp(ref)
if SELECTION == "prop":
values = df_prop["feedback_accuracy"]
values_ref = df_prop_ref["feedback_accuracy"]
else:
values = df_resp["feedback_fairness"]
values_ref = df_resp_ref["feedback_fairness"]
# feedback_fairness
values_ref = values_ref.apply(lambda x: AI_FEEDBACK_ACCURACY_SCALAS_REV.get(x, x))
values = values.apply(lambda x: AI_FEEDBACK_ACCURACY_SCALAS_REV.get(x, x))
# print("DIFF: ", values, values_ref)
# resp_values = metrics.get_data(metrics.get_rel_min_offer_df(df_resp))
# resp_ref_values = metrics.get_data(metrics.get_rel_min_offer_df(df_resp_ref))
values
print("MEDIAN: ", values.median(), values_ref.median())
dof = 0
diff = 0
table, res = rp.crosstab(pd.Series(values), pd.Series(values_ref), test='g-test')
s, p, r = res.results.values
# s = res.results[2]
# p = res.results[3]
# r = res.results[9]
# diff = res.results[0]
# dof = res.results[1]
# s = res.results[2]
# p = res.results[3]
# r = res.results[9]
tmp_res = None
tmp_res = stats.mannwhitneyu(values, values_ref, use_continuity=False)
# tmp_res = stats.ranksums(values, values_ref)
print("TMP values: ", tmp_res)
print("Conclusion: ", generate_stat_sentence(np.mean(values_ref), np.std(values_ref), np.mean(values), np.std(values), s, p, dof, diff=diff, label1="t12.dss", label2=treatment+".dss"))
print("Table:", table)
print("Res:", res)
res = {
"rel. min_offer T12": metrics.get_mean(values_ref),
"rel. min_offer T13": metrics.get_mean(values),
# "rejection_ratio": rejection_ratio(df_prop)
}
test_label = f"(ttest independent) H0: equal"
res = {k: (f"{v:.3f}" if pd.notnull(v) and v!= int(v) else v) for k,v in res.items()}
res["min_offer" + test_label] = f"{s:.3f} (p: {p:.3f}, r: {r:.3f})"
return res
statDF['ShowCommitmentPercent'] = statDF.apply(
lambda x: 1 - x['avoidCommitPercent'], axis=1)
statDF = statDF.reset_index()
statDF['participantsType'] = [
'RL Agent' if 'max' in name else 'Human' for name in statDF['name']
]
# statDF['sem'] = df.groupby(['participantsType', 'decisionSteps'])["avoidCommitPercent"].apply(calculateSE)
statDF = statDF[statDF['participantsType'] == 'Human']
# statDF = statDF[statDF['participantsType'] == 'RL Agent']
crosstab, res = researchpy.crosstab(dfExpTrail['hasAvoidPoint'],
dfExpTrail['decisionSteps'],
test="chi-square")
print(crosstab)
# Compute the two-way mixed-design ANOVA
calAnova = 0
if calAnova:
import pingouin as pg
aov = pg.mixed_anova(dv='ShowCommitmentPercent',
within='decisionSteps',
between='participantsType',
subject='name',
data=statDF)
pg.print_table(aov)
def _get_prop_vs_prop_dss_score(treatment,
con,
dfs=None,
use_percentage=None,
use_labels=None,
metric=None,
as_percentage=None,
is_categorical=None,
alternative=None):
df_prop, df_resp = get_prop_resp(treatment)
df_prop_ref, df_resp_ref = get_prop_resp("t11a")
print(metric.__name__)
metric_values = metric(df_prop)
metric_value = metrics.get_mean(metric_values)
metric_ref_values = metric(df_prop_ref)
metric_value_ref = metrics.get_mean(metric_ref_values)
metric_values = metrics.get_data(metric_values)
metric_ref_values = metrics.get_data(metric_ref_values)
dof = 0
diff = None
if is_categorical:
table, res = rp.crosstab(pd.Series(metric_ref_values),
pd.Series(metric_values),
test='chi-square')
s, p, r = res.results.values
print(
"Conclusion: ",
generate_cat_stat_sentence(np.mean(metric_ref_values),
np.std(metric_ref_values),
np.mean(metric_values),
np.std(metric_values),
s,
p,
dof,
diff=diff,
label1="t11a.dss",
label2=treatment + ".dss"))
test_label = f"(pearson chi2)"
else:
#print("Ranksums", stats.ranksums(metric_ref_values, metric_values))
table, res = rp.ttest(pd.Series(metric_ref_values),
pd.Series(metric_values),
paired=False)
s = res.results[2]
if alternative == "greater":
p = res.results[4]
elif alternative == "less":
p = res.results[5]
elif alternative in (None, 'two-sided'):
p = res.results[3]
r = res.results[9]
diff = res.results[0]
dof = res.results[1]
s = res.results[2]
p = res.results[3]
r = res.results[9]
print(
"Conclusion: ",
generate_stat_sentence(np.mean(metric_ref_values),
np.std(metric_ref_values),
np.mean(metric_values),
np.std(metric_values),
s,
p,
dof,
diff=diff,
label1="t11a.dss",
label2=treatment + ".dss"))
test_label = f"(ttest independent) H0: {'equal' if alternative in {None, 'two-sided'} else alternative}"
print("RESUME: ", res)
print("TABLE: ", table)
if as_percentage:
res = {
# "Proposer": f'{100 * prop_value:.2f} %',
"Proposer + DSS":
f'{100 * metric_value:.2f} %',
"T11A ":
f'{100 * metric_value_ref:.2f} %',
"prop:dss - auto prop":
f'{100 * (metric_value - metric_value_ref):.2f} %',
}
else:
res = {
# "Proposer": f'{prop_value:.2f}',
"Proposer + DSS": f'{metric_value:.2f}',
"T11A": f'{metric_value_ref:.2f}',
"prop:dss - auto prop":
f'{(metric_value - metric_value_ref):.2f} %',
}
res[test_label] = f"{s:.3f} ({p:.3f})"
if is_categorical:
res[test_label] = f"{s:.3f} (p: {p:.3f}, phi: {r:.3f})"
else:
res[test_label] = f"{s:.3f} (p: {p:.3f}, r: {r:.3f})"
return res
plt.ylabel('PDF(IMDB_Class Rating)')
plt_params_x = imdb_db.class_rating.value_counts().index
plt_params_y = imdb_db.class_rating.value_counts().values
sb.barplot(plt_params_x, plt_params_y)
imdb_db.class_rating.value_counts()
# In[35]:
# sb.barplot(imdb_db.class_rating.values, hue='color', data=imdb_db)
# In[36]:
import researchpy as rp
from scipy import stats
table, results, expected = rp.crosstab(imdb_db['color'],
imdb_db['class_rating'],
test='chi-square',
expected_freqs=True)
table, results
#Just to verify with Chi-square test of independance if the categorical values 'color' and 'black and white' are independance.
#p-value of <=0.05 indicates, there is indeed independance. So we keep the attribute 'color' and move ahead
# In[37]:
imdb_db.columns
# In[38]:
# import researchpy as rp
# from scipy import stats
# table, results, expected = rp.crosstab(imdb_db['aspect_ratio'], imdb_db['class_rating'], test= 'chi-square', expected_freqs=True)
# table , results
qual = 'Small'
elif etaSq < .14:
qual = 'Medium'
else:
qual = 'Large'
return (qual)
print(" Age 和 Survived 是否有相關性?", judgment_etaSq(etaSq))
# Case02_透過數值法計算 Sex 和 Survived_cate 是否有相關性?
contTable = pd.crosstab(df_train['Sex'], df_train['Survived_cate'])
print(contTable)
df = min(contTable.shape[0], contTable.shape[1]) - 1
crosstab, res = researchpy.crosstab(df_train['Sex'],
df_train['Survived_cate'],
test='chi-square')
print("Cramer's value is", res.loc[2, 'results'])
## 寫一個副程式判斷相關性的強度
def judgment_CramerV(df, V):
if df == 1:
if V < 0.10:
qual = 'negligible'
elif V < 0.30:
qual = 'small'
elif V < 0.50:
qual = 'medium'
else:
qual = 'large'
