def sign_barplot(df, val_col, group_col, test="HSD"):
if test == "HSD":
result_df = tukey_hsd(df, val_col, group_col)
if test == "tukey":
result_df = sp.posthoc_tukey(df, val_col, group_col)
if test == "ttest":
result_df = sp.posthoc_ttest(df, val_col, group_col)
if test == "scheffe":
result_df = sp.posthoc_scheffe(df, val_col, group_col)
if test == "dscf":
result_df = sp.posthoc_dscf(df, val_col, group_col)
if test == "conover":
result_df = sp.posthoc_conover(df, val_col, group_col)
#マッピングのプロファイル
fig, ax = plt.subplots(1, 2, figsize=(10, 6))
cmap = ['1', '#fb6a4a', '#08306b', '#4292c6', '#c6dbef']
heatmap_args = {
'cmap': cmap,
'linewidths': 0.25,
'linecolor': '0.5',
'clip_on': False,
'square': True
}
sp.sign_plot(result_df, ax=ax[1], **heatmap_args) #検定結果を描画
sns.barplot(data=df, x=group_col, y=val_col, capsize=0.1,
ax=ax[0]) #使ったデータを描画
plt.show()
print(p_value)
#Hypothesis that the variances are equal can't be rejected and each sample has more than 5 individuals
#(p_value=0.28>0.05 ==> H0 failed to be rejected)
#All assumptions for Kruskal are respected
from scipy.stats import kruskal
H, p_value = kruskal(BECH_mean, SW_mean, SL_mean)
print(H)
print(p_value)
#p_value=0.002, it means the samples have different medians
data_dict = {'BECH': BECH_mean, 'SW': SW_mean, 'SL': SL_mean}
df = pd.DataFrame({k: pd.Series(v) for k, v in data_dict.items()})
df = df.melt(var_name='groups', value_name='values')
from scikit_posthocs import posthoc_dscf
posthoc_dscf(df, val_col='values', group_col='groups')
#output shows each sample is different from the others:
# BECH SL SW
#BECH -1.000 0.001000 0.001000
#SL 0.001 -1.000000 0.009617
#SW 0.001 0.009617 -1.000000
#paired p_values are all smaller than 0.05, so each sample is significantly different from the
#two others
import pylab
BoxName = ['BECH', 'SL', 'SW']
data = [BECH_mean, SL_mean, SW_mean]
pylab.xticks([1, 2, 3], BoxName)
plt.boxplot(data)
plt.savefig('MultipleBoxPlot.png')
plt.show()
################################################################
print('counter variance')
################################################################
summary_df["cv"] = summary_df.std_vel / summary_df.mean_vel
_, norm_p = stats.shapiro(summary_df.cv.dropna())
_, var_p = stats.levene(
summary_df[summary_df.experiment_type == 'baseline'].cv.dropna(),
summary_df[summary_df.experiment_type == 'control'].cv.dropna(),
summary_df[summary_df.experiment_type == 'button'].cv.dropna(),
summary_df[summary_df.experiment_type == 'touch'].cv.dropna(),
center='median'
)
if norm_p < 0.05 or var_p < 0.05:
print('steel-dwass\n', sp.posthoc_dscf(summary_df, val_col='cv', group_col='experiment_type'))
else:
multicomp_result = multicomp.MultiComparison(np.array(summary_df.dropna(how='any').cv, dtype="float64"), summary_df.dropna(how='any').experiment_type)
print('levene', multicomp_result.tukeyhsd().summary())
print ('----', 'baseline', 'control', 'button', 'touch')
print('mean',
summary_df[summary_df.experiment_type == 'baseline'].cv.dropna().mean(),
summary_df[summary_df.experiment_type == 'control'].cv.dropna().mean(),
summary_df[summary_df.experiment_type == 'button'].cv.dropna().mean(),
summary_df[summary_df.experiment_type == 'touch'].cv.dropna().mean(),
)
print('var',
summary_df[summary_df.experiment_type == 'baseline'].cv.dropna().std(),
summary_df[summary_df.experiment_type == 'control'].cv.dropna().std(),
summary_df[summary_df.experiment_type == 'button'].cv.dropna().std(),
summary_df[summary_df.experiment_type == 'touch'].cv.dropna().std(),
def show_statistics(self, feature_type):
print("Focus on number of {}".format(feature_type))
feature_df = pd.DataFrame(
[data[feature_type] for data in self.feature.values()],
index=self.categories.values()).T
fig, ax = plt.subplots(figsize=(8, 4))
ax.boxplot(feature_df.values, 0, "")
ax.set_xticklabels(self.categories_en.values())
plt.title('Box plot')
plt.xlabel('category')
plt.ylabel('number of {}'.format(feature_type))
plt.show()
print("========================================"
"========================================")
print("要約統計量")
display(feature_df.describe())
print("========================================"
"========================================")
print("データの正規性を調べる")
print("Kolmogorov–Smirnov検定")
ks_null_hypothesis = "データの分布は正規分布と差がない"
ks_alternative_hypothesis = "データの分布は正規分布と差がある"
print("帰無仮説: {}".format(ks_null_hypothesis))
print("対立仮説: {}".format(ks_alternative_hypothesis))
ks_a_percentage = self.ks_a_percentage
print("有意水準: {}%（{}）".format(ks_a_percentage, ks_a_percentage * 0.01))
ks_p_values = [
st.kstest(data, "norm")[1] for data in feature_df.values.T
]
print("p値")
for cat, p_value in zip(self.categories.values(), ks_p_values):
print("{}: {}".format(cat, p_value))
if np.min(ks_p_values) < ks_a_percentage * 0.01:
print("いずれかの群のデータに於いて，"
"帰無仮説が棄却されたため「{}」とは言えない，つまり「{}」".format(
ks_null_hypothesis, ks_alternative_hypothesis))
print("=> ノンパラメトリック検定へ")
print("======================================"
"==========================================")
print("カテゴリ間に差があるか調べる")
print("Kruskal-Wallis検定")
kruskal_null_hypothesis = "カテゴリ間で代表値（中央値）に差はない"
kruskal_alternative_hypothesis = "いずれかのカテゴリ間で"\
"代表値（中央値）に差がある"
print("帰無仮説: {}".format(kruskal_null_hypothesis))
print("対立仮説: {}".format(kruskal_alternative_hypothesis))
kruskal_a_percentage = self.kruskal_a_percentage
print("有意水準: {}%（{}）".format(kruskal_a_percentage,
kruskal_a_percentage * 0.01))
kruskal_p_value = st.kruskal(
*[data for data in feature_df.values.T])[1]
print("p値: {}".format(kruskal_p_value))
if kruskal_p_value < kruskal_a_percentage * 0.01:
print("帰無仮説が棄却されたため，「{}」とは言えない，つまり「{}」".format(
kruskal_null_hypothesis, kruskal_alternative_hypothesis))
print("======================================"
"==========================================")
print("多重比較")
print("Steel-Dwass検定")
steel_dwass_null_hypothesis = "2つのカテゴリ間の"\
"代表値（中央値）に差はない"
print("帰無仮説: {}".format(steel_dwass_null_hypothesis))
steel_dwass_a_percentage = self.steel_dwass_a_percentage
print("優位水準: {}%（{}）".format(steel_dwass_a_percentage,
steel_dwass_a_percentage * 0.01))
values = []
groups = []
categories = self.categories
for data, cat in zip(feature_df.T.values, categories.values()):
groups.extend([cat for i in range(len(data))])
values.extend(data)
posthoc_df = pd.DataFrame([values, groups],
index=["value", "category"]).T
print("以下p値の表")
display(
sp.posthoc_dscf(posthoc_df,
val_col="value",
group_col="category"))
else:
print("帰無仮説が棄却されなかったため，"
"「{}」は誤っているとは言えない".format(kruskal_null_hypothesis))
print("\n")
def steel_dwass(df, x, y, **kwargs):
"""
TukeyHSD のノンパラメトリック版
"""
return sp.posthoc_dscf(df, val_col=y, group_col=x, **kwargs)