def friedman_posthoc_tests(experiment_pivot_df):
"""Returns p-value tables for various Friedman posthoc tests.
Results should considered only if Friedman test rejects null hypothesis.
"""
posthoc_tests = {}
posthoc_tests['conover'] = sp.posthoc_conover_friedman(experiment_pivot_df)
posthoc_tests['nemenyi'] = sp.posthoc_nemenyi_friedman(experiment_pivot_df)
return posthoc_tests
_, norm_p2 = stats.shapiro(inttype_accuracy.CONTROL)
_, norm_p3 = stats.shapiro(inttype_accuracy.BUTTON)
_, norm_p4 = stats.shapiro(inttype_accuracy.TOUCH)
_, var_p = stats.levene(inttype_accuracy.CONTROL,
inttype_accuracy.BUTTON,
inttype_accuracy.TOUCH,
center='median')
if norm_p2 < 0.05 or norm_p3 < 0.05 or norm_p4 < 0.05:
_, anova_p = stats.friedmanchisquare(inttype_accuracy.CONTROL,
inttype_accuracy.BUTTON,
inttype_accuracy.TOUCH)
if anova_p < 0.05:
print("conover test anova_result:", anova_p,
sp.posthoc_conover_friedman(inttype_accuracy))
else:
melted_df = pd.melt(inttype_accuracy.reset_index(),
id_vars="subject",
var_name="experiment_type",
value_name="accuracy")
anova_result = stats_anova.AnovaRM(melted_df, "accuracy", "subject",
["experiment_type"])
print("reperted anova: ", anova_result.fit())
melted_df = pd.melt(inttype_accuracy,
var_name="experiment_type",
value_name="accuracy")
print("levene result", var_p)
# gamesHowellTest(melted_df, "experiment_type", "accuracy")
multicomp_result = multicomp.MultiComparison(
np.array(melted_df.dropna(how='any').accuracy, dtype="float64"),
]:
print("Testing using different OWA weights")
results[emo].append(
main(data_frame=dat,
mlm_method=current_methods[emo](owa),
evaluation=5)[0])
results[emo].append(
main(data_frame=dat, mlm_method=recomended_methods[emo],
evaluation=5)[0])
resultsdf = pd.DataFrame(results)
friedman = st.friedmanchisquare(*resultsdf.values)
print("The statistic for FRNN")
print(friedman)
if friedman.pvalue < 0.05:
p_values = hocs.posthoc_conover_friedman(resultsdf.T, p_adjust="holm")
p_values.to_excel(r"Final Output\results_OWA_operators_FRNN_p_values.xlsx")
ranks = pd.DataFrame(columns=resultsdf.keys())
for key in resultsdf.keys():
ranks[key] = resultsdf[key].rank(ascending=False)
resultsdf["Totals"] = resultsdf.sum(axis=1) / 4
resultsdf["Ranks"] = ranks.mean(axis=1)
resultsdf["Nearest Neighbours"] = [
"Strict", "Invadd", "Additive", "Exponential", "Mean", "Trimmed",
"Recommended"
]
resultsdf.to_excel(r"Final Output\results_OWA_operators_FRNN.xlsx")
##
# For the baseline methods determine optimal k, using the optimal weighting schemes for FRNN per dataset.
methods = [
medias = df2.mean()
aver1 = [medias[0], medias[2], medias[4], medias[6], medias[8]]
aver2 = [medias[1], medias[3], medias[5], medias[7], medias[9]]
# plt.plot([5,10,15,20,25], aver1, [5,10,15,20,25], aver2,)
print(medias)
# sns.boxplot(data=df, x="Cases", y="TTotal", hue=df.Cases.tolist())
fig, ax = plt.subplots(1, 1, figsize=(10, 5))
part = [
23.035425186157227, 9.456007719039917, 8.742131471633911,
10.28268098831176, 11.558128356933594
]
# ax.plot([5,10,15,20,25], aver1, marker='o')
ax.plot([5, 10, 15, 20, 25], part, marker='o')
ax.set_xlabel("K", fontsize=15)
ax.set_ylabel("sec.", fontsize=15)
ax.xaxis.set_major_locator(mpl.ticker.FixedLocator([5, 10, 15, 20, 25]))
plt.show()
print(pg.friedman(data=df2))
t = sp.posthoc_conover_friedman(a=df2)
print(t)
print("Task1 Completion Time:\n", task1_time_avg)
print("Task1 Completion Time 95\% CI:\n", task1_time_ci)
task1_acc_detail = [[
np.average(task1_acc[:, i, k, j]) for i in range(8) for j in range(2)
] for k in range(7)]
task1_time_detail = [[
np.average(task1_time[:, i, k, j]) for i in range(8) for j in range(2)
] for k in range(7)]
acc_fm_res = stats.friedmanchisquare(*task1_acc_detail)
time_fm_res = stats.friedmanchisquare(*task1_time_detail)
print(acc_fm_res)
print(
sp.posthoc_conover_friedman(np.array(task1_acc_detail).T,
p_adjust=None))
print(time_fm_res)
print(sp.posthoc_conover_friedman(np.array(task1_time_detail).T))
print()
print("E2")
# Task 2 - density_2
task2_acc = np.zeros((total_obtain_answers, 8, 7, 2)).astype(np.int)
task2_time = np.zeros((total_obtain_answers, 8, 7, 2)).astype(np.float)
learning_effect_acc = np.zeros(
(total_obtain_answers, 8, 14)).astype(np.float)
learning_effect_time = np.zeros(
(total_obtain_answers, 8, 14)).astype(np.float)
task2_personal = np.zeros((total_obtain_answers)).astype(np.float)
offset = 2 + 9 * 2 * 7 + 20
# inttype_accuracy = inttype_accuracy.append(buf, ignore_index=True)
inttype_accuracy.mean()
inttype_accuracy.std()
# inttype_accuracy_cross = pd.crosstab(inttype_accuracy.experiment, inttype_accuracy.result)
# stats.chi2_contingency(inttype_accuracy_cross)
_, norm_p1 = stats.shapiro(inttype_accuracy.BASELINE)
_, norm_p2 = stats.shapiro(inttype_accuracy.CONTROL)
_, norm_p3 = stats.shapiro(inttype_accuracy.BUTTON)
_, norm_p4 = stats.shapiro(inttype_accuracy.TOUCH)
_, var_p = stats.levene(inttype_accuracy.BASELINE, inttype_accuracy.CONTROL, inttype_accuracy.BUTTON, inttype_accuracy.TOUCH, center='median')
if norm_p1 < 0.05 or norm_p2 < 0.05 or norm_p3 < 0.05 or norm_p4 < 0.05:
_, anova_p = stats.friedmanchisquare(inttype_accuracy.BASELINE, inttype_accuracy.CONTROL, inttype_accuracy.BUTTON, inttype_accuracy.TOUCH)
if anova_p < 0.05:
print("conover test anova_result:", anova_p, sp.posthoc_conover_friedman(inttype_accuracy))
else:
melted_df = pd.melt(inttype_accuracy.reset_index(), id_vars="subject", var_name="experiment_type", value_name="accuracy")
anova_result = stats_anova.AnovaRM(melted_df, "accuracy", "subject", ["experiment_type"])
print("reperted anova: ", anova_result.fit())
melted_df = pd.melt(inttype_accuracy, var_name="experiment_type", value_name="accuracy")
print("levene result", var_p)
# gamesHowellTest(melted_df, "experiment_type", "accuracy")
multicomp_result = multicomp.MultiComparison(np.array(melted_df.dropna(how='any').accuracy, dtype="float64"), melted_df.dropna(how='any').experiment_type)
print(multicomp_result.tukeyhsd().summary())
subject_accuracy = inttype_accuracy.T
_, anova_p = stats.friedmanchisquare(
subject_accuracy.ando,
subject_accuracy.aso,
subject_accuracy.hikosaka,