def test_oneway(self):
r1 = self.res_oneway
r2s = self.res_2s
res_bfm = self.res_bfm
res_wa = self.res_wa
res_fa = self.res_fa
# check we got the correct data
m = [x_i.mean() for x_i in self.x]
assert_allclose(m, self.res_m, rtol=1e-13)
# 3 sample case
resg = smo.anova_oneway(self.x, use_var="unequal", trim_frac=1 / 13)
# assert_allclose(res.statistic, res_bfm.statistic, rtol=1e-13)
assert_allclose(resg.pvalue, r1.p_value, rtol=1e-13)
assert_allclose(resg.df, [r1.df1, r1.df2], rtol=1e-13) # df
# 2-sample against yuen t-test
resg = smo.anova_oneway(self.x[:2],
use_var="unequal",
trim_frac=1 / 13)
# assert_allclose(res.statistic, res_bfm.statistic, rtol=1e-13)
assert_allclose(resg.pvalue, r2s.p_value, rtol=1e-13)
assert_allclose(resg.df, [1, r2s.df], rtol=1e-13) # df
# no trimming bfm
res = smo.anova_oneway(self.x, use_var="bf")
assert_allclose(res[0], res_bfm.statistic, rtol=1e-13)
assert_allclose(res.pvalue2, res_bfm.p_value, rtol=1e-13)
assert_allclose(res.df2, res_bfm.parameter, rtol=1e-13) # df
# no trimming welch
res = smo.anova_oneway(self.x, use_var="unequal")
assert_allclose(res.statistic, res_wa.statistic, rtol=1e-13)
assert_allclose(res.pvalue, res_wa.p_value, rtol=1e-13)
assert_allclose(res.df, res_wa.parameter, rtol=1e-13) # df
# no trimming standard anova
res = smo.anova_oneway(self.x, use_var="equal")
assert_allclose(res.statistic, res_fa.statistic, rtol=1e-13)
assert_allclose(res.pvalue, res_fa.p_value, rtol=1e-13)
assert_allclose(res.df, res_fa.parameter, rtol=1e-13) # df
def test_means(self):
# library onewaystats, BF test for equality of means
# st = bf.test(y ~ g, df3)
statistic = 7.10900606421182
parameter = [2, 31.4207256105052]
p_value = 0.00283841965791224
# method = 'Brown-Forsythe Test'
res = anova_oneway(self.data, use_var="bf")
# R bf.test uses original BF df_num
assert_allclose(res.pvalue2, p_value, rtol=1e-13)
assert_allclose(res.statistic, statistic, rtol=1e-13)
assert_allclose([res.df_num2, res.df_denom], parameter)
def test_options(self):
# regression tests for options,
# many might not be implemented in other packages
data = self.data
# regression numbers from initial run
statistic, p_value = 1.0173464626246675, 0.3763806150460239
df = (2.0, 24.40374758005409)
res = smo.test_scale_oneway(data,
method='unequal',
center='median',
transform='abs',
trim_frac_mean=0.2)
assert_allclose(res.pvalue, p_value, rtol=1e-13)
assert_allclose(res.statistic, statistic, rtol=1e-13)
assert_allclose(res.df, df)
statistic, p_value = 1.0329722145270606, 0.3622778213868562
df = (1.83153791573948, 30.6733640949525)
p_value2 = 0.3679999679787619
df2 = (2, 30.6733640949525)
res = smo.test_scale_oneway(data,
method='bf',
center='median',
transform='abs',
trim_frac_mean=0.2)
assert_allclose(res.pvalue, p_value, rtol=1e-13)
assert_allclose(res.statistic, statistic, rtol=1e-13)
assert_allclose(res.df, df)
assert_allclose(res.pvalue2, p_value2, rtol=1e-13)
assert_allclose(res.df2, df2)
statistic, p_value = 1.7252431333701745, 0.19112038168209514
df = (2.0, 40.0)
res = smo.test_scale_oneway(data,
method='equal',
center='mean',
transform='square',
trim_frac_mean=0.2)
assert_allclose(res.pvalue, p_value, rtol=1e-13)
assert_allclose(res.statistic, statistic, rtol=1e-13)
assert_equal(res.df, df)
statistic, p_value = 0.4129696057329463, 0.6644711582864451
df = (2.0, 40.0)
res = smo.test_scale_oneway(
data,
method='equal',
center='mean',
transform=lambda x: np.log(x * x), # noqa
trim_frac_mean=0.2)
assert_allclose(res.pvalue, p_value, rtol=1e-13)
assert_allclose(res.statistic, statistic, rtol=1e-13)
assert_allclose(res.df, df)
# compare no transform with standard anova
res = smo.test_scale_oneway(data,
method='unequal',
center=0,
transform='identity',
trim_frac_mean=0.2)
res2 = anova_oneway(self.data, use_var="unequal")
assert_allclose(res.pvalue, res2.pvalue, rtol=1e-13)
assert_allclose(res.statistic, res2.statistic, rtol=1e-13)
assert_allclose(res.df, res2.df)
print(sm.stats.anova_lm(st_vfcc, typ=2))
print('Residual Shapiro ' + str(stats.shapiro(st_vfcc.resid)))
print(stats.anderson(st_vfcc.resid))
mm.probability_plot_ols(10, st_vfcc.resid)
#except Exception:
# pass
try:
print('')
print('Welch ANOVA')
print('')
welch = ao.anova_oneway(df['vF'],
df['Depth'],
use_var='unequal',
welch_correction=True)
print('-------- vF ----------')
print(welch)
welch = ao.anova_oneway(df['vwc1kPa'],
df['Depth'],
use_var='unequal',
welch_correction=True)
print('-------- vwc1kPa ----------')
print(welch)
welch = ao.anova_oneway(df['vwc3kPa'],
df['Depth'],
use_var='unequal',
welch_correction=True)
bartlett = stats.bartlett(
df[df['indp_var']=='grad'][col],
df[df['indp_var']=='wish'][col],
df[df['indp_var']=='no_wish'][col]
)
if not levene[1] > 0.05 or bartlett[1] > 0.05:
noequal_var.append(col)
else:
print(col)
# Brown-Forsythe test for no equality of varience
brown = []
for col in noequal_var:
brown.append(
anova_oneway(
df[col], df['indp_var'],
use_var='bf'
).pvalue2)
no_equal_df = pd.DataFrame([noequal_var, brown]).T
no_equal_df.to_csv(data_loc + 'no_equalANOVA.csv', encoding='cp949')
# Only one column satisfy an equality of varience
model = ols('교수수업만족 ~ C(indp_var)', df).fit()
print(anova_lm(model))
anova_lm(model).to_csv(data_loc + 'equl_교수수업만족anova.csv')
# Chi-square for categorical y variables
chi_score = []
p_value = []
degree_f = []
for col in categorical:
