def IQR(regions=None, countries=None):
"""IQR method for region dataframe.
Args:
regions (pd.DataFrame, optional): Optional data to work on.
If not given, all regions used.
countries ()
"""
# fetch region dataframe
regions = _src.regions_df() if regions is None else regions
# subset country
if countries:
regions = regions[regions.Country.isin(list(countries))]
descriptive_cols = {'Region', 'Code', 'Country'}
def _IQR(x):
"""IQR method for a vector."""
q1, q3 = np.percentile(sorted(x), [25, 75])
iqr = q3 - q1
low, high = (q1 - (1.5 * iqr), q3 + (1.5 * iqr))
return (x < low) | (x > high)
# remove descriptive columns
outs = {}
for c in set(regions.columns) - descriptive_cols:
flags = _IQR(regions[c])
outs[c] = regions[flags]
# return outliers
return outs
def popdensity_boxplot_noPRG(name=None, regions_df=None):
# default data if not given
regions_df = regions_df if regions_df is not None else _src.regions_df()
# crop prague off
regions_df = regions_df[regions_df.Code != "CZ010"]
# plot
popdensity_boxplot(name=name, regions_df=regions_df)
def administrative_divisions_similar(pi=True, alpha=.05):
"""Tests that regions have same mean in their populations, areas and densities.
Use two-sampled t_test with preceding F-test to test equal variances.
Only the result of t_test is returned.
H0: mu1 = mu2
HA: mu1 != mu2
Args:
pi (bool): If True, returns pi values.
If False, returns validity of H0.
Defautly True.
alpha (float): Significance level.
"""
# data
attributes = ['Population', 'Area', 'Density']
regions_df = _src.regions_df()
pi_df = _src.regions_countries_pairs(attributes)
# perform the test
for a in attributes:
for i, r in pi_df.iterrows():
# data
data1 = regions_df[regions_df.Country == r.Country1][a]
data2 = regions_df[regions_df.Country == r.Country2][a]
# F-test
#ftest, fpi = _tools.f_test(data1, data2)
ftest, fpi = stats.levene(data1, data2)
equal_var = fpi > alpha
# test
pop_pi = stats.ttest_ind(data1, data2, equal_var=equal_var)
# write down pvalue
pi_df.at[i, a] = pop_pi.pvalue
# return pi value
if pi: return pi_df
# make decision
pi_df = pd.concat(
[pi_df[['Country1', 'Country2']], pi_df[attributes] > alpha],
axis=1,
ignore_index=True)
pi_df.columns = ['Country1', 'Country2', *attributes]
return pi_df
def regions_normally_distributed(pi=True, alpha=.05):
"""Tests whether regions are distributed normally in their
populations, areas and densities over different countries.
Uses Shapiro-Wilk test.
H0: Regions of country are distributed normally in the attribute.
HA: They are not.
Args:
pi (bool): If True, returns pi values.
If False, returns validity of H0.
Defautly True.
alpha (float): Significance level.
"""
# data
attributes = ['Population', 'Area', 'Density']
regions_df = _src.regions_df()
# empty single country
countries = regions_df.Country.unique()
# create dataframe
pi_dict = {k: [None for _ in range(len(countries))] for k in attributes}
pi_dict = {'Country': [c for c in countries], **pi_dict}
# pi values dataframes
pi_df = pd.DataFrame(pi_dict)
# perform the test
for a in attributes:
for i, r in pi_df.iterrows():
# data
data = regions_df[regions_df.Country == r.Country][a]
# Shapiro-Wilk test
normal_pi = stats.shapiro(data)
pi_df.at[i, a] = normal_pi.pvalue
# return pi value
if pi: return pi_df
# make decision
pi_df = pd.concat([pi_df[['Country']], pi_df[attributes] > alpha],
axis=1,
ignore_index=True)
pi_df.columns = ['Country', *attributes]
return pi_df
def area_population_scatter(name=None, regions_df=None):
# default data if not given
regions_df = regions_df if regions_df is not None else _src.regions_df()
# plot
ax = sns.jointplot(x="Area",
y="Population",
hue="Country",
data=regions_df)
# axis limits
ax.ax_marg_x.set_xlim(-1.5 * 10**4, 1.25 * 10**5)
ax.ax_marg_y.set_ylim(-5 * 10**5, 6 * 10**6)
if name is None:
plt.show()
# save
else:
plt.savefig(name)
def popdensity_boxplot(name=None, regions_df=None):
# default data if not given
regions_df = regions_df if regions_df is not None else _src.regions_df()
# plot
plt.rcParams.update({'font.size': 20})
#plt.yscale("log")
sns.violinplot(x="Country", y="Density", data=regions_df)
#sns.boxplot(x="Country", y="Density", data=regions_df, color = "1")
sns.stripplot(x="Country",
y="Density",
color='black',
size=6,
alpha=0.8,
data=regions_df)
if name is None:
plt.show()
# save
else:
plt.savefig("density.png")