def generate_main_results():
"""Generate the main results of the experiment."""
wide_optimal = sort_tbl(
calculate_wide_optimal(results),
ovrs_order=OVRS_NAMES, clfs_order=CLFS_NAMES
)\
.set_index(['Dataset', 'Classifier', 'Metric'])\
.apply(lambda row: make_bold(row, num_decimals=3), axis=1)\
.reset_index()
wide_optimal['Dataset'] = wide_optimal['Dataset'].apply(
lambda x: x.title()
if len(x.split(' ')) == 1
else ''.join([w[0] for w in x.split(' ')])
)
mean_sem_scores = sort_tbl(
generate_mean_std_tbl_bold(*calculate_mean_sem_scores(results), maximum=True, decimals=3),
ovrs_order=OVRS_NAMES, clfs_order=CLFS_NAMES
)
mean_sem_perc_diff_scores = sort_tbl(
generate_mean_std_tbl(*calculate_mean_sem_perc_diff_scores(results, ['SMOTE', 'K-SMOTE'])),
ovrs_order=OVRS_NAMES, clfs_order=CLFS_NAMES
)
mean_sem_ranking = sort_tbl(
generate_mean_std_tbl_bold(*calculate_mean_sem_ranking(results), maximum=False),
ovrs_order=OVRS_NAMES, clfs_order=CLFS_NAMES
)
main_results_names = ('wide_optimal', 'mean_sem_scores', 'mean_sem_perc_diff_scores', 'mean_sem_ranking')
return zip(main_results_names, (wide_optimal, mean_sem_scores, mean_sem_perc_diff_scores, mean_sem_ranking))
def generate_main_results():
"""Generate the main results of the experiment."""
main_results = {}
for ratio in UNDERSAMPLING_RATIOS:
# Generate results
results = generate_results(ratio)
# Calculate results
mean_sem_scores = sort_tbl(
generate_mean_std_tbl(*calculate_mean_sem_scores(results)),
ovrs_order=OVERSAMPLERS_NAMES,
clfs_order=CLASSIFIERS_NAMES)
mean_sem_perc_diff_scores = sort_tbl(
generate_mean_std_tbl(*calculate_mean_sem_perc_diff_scores(
results, ['NO OVERSAMPLING', 'G-SMOTE'])),
ovrs_order=OVERSAMPLERS_NAMES,
clfs_order=CLASSIFIERS_NAMES)
mean_sem_ranking = sort_tbl(
generate_mean_std_tbl(*calculate_mean_sem_ranking(results)),
ovrs_order=OVERSAMPLERS_NAMES,
clfs_order=CLASSIFIERS_NAMES)
# Populate main results
main_results_names = ('mean_sem_scores', 'mean_sem_perc_diff_scores',
'mean_sem_ranking')
main_results[ratio] = zip(
main_results_names,
(mean_sem_scores, mean_sem_perc_diff_scores, mean_sem_ranking))
return main_results
def test_mean_sem_ranking():
"""Test the mean ranking results of experiment."""
mean_ranking, sem_ranking = calculate_mean_sem_ranking(EXPERIMENT.results_)
assert set(mean_ranking.Classifier.unique()) == set(
EXPERIMENT.classifiers_names_)
assert set(sem_ranking.Classifier.unique()) == set(
EXPERIMENT.classifiers_names_)
assert set(mean_ranking.columns[2:]) == set(EXPERIMENT.oversamplers_names_)
assert set(sem_ranking.columns[2:]) == set(EXPERIMENT.oversamplers_names_)
assert len(mean_ranking) == len(CLASSIFIERS)
assert len(sem_ranking) == len(CLASSIFIERS)
def make_mean_rank_bar_chart():
"""Generates bar chart."""
load_plt_sns_configs()
ranks = calculate_mean_sem_ranking(results)[0]
ranks['Metric'] = ranks['Metric'].apply(lambda x: METRICS_MAPPING[x])
fig, axes = plt.subplots(
ranks.Classifier.unique().shape[0],
ranks.Metric.unique().shape[0],
figsize=(5,6)
)
lranks = ranks.set_index(['Classifier', 'Metric'])
for (row, clf), (col, metric) in product(
enumerate(ranks.Classifier.unique()),
enumerate(ranks.Metric.unique())
):
dat = len(OVRS_NAMES) - lranks.loc[(clf,metric)].loc[list(OVRS_NAMES[::-1])]
axes[row, col].bar(
dat.index,
dat.values,
color=['indianred']+['steelblue' for i in range(len(OVRS_NAMES)-1)]
)
plt.sca(axes[row, col])
plt.yticks(range(len(OVRS_NAMES)), [None]+list(range(1, len(OVRS_NAMES)))[::-1])
plt.xticks(rotation=90)
if row == 0:
plt.title(metric)
if col == 0:
plt.ylabel(f'{clf}')
if row != len(ranks.Classifier.unique())-1:
plt.xticks(range(len(OVRS_NAMES)), [])
if col != 0:
plt.yticks(range(len(OVRS_NAMES)), [])
sns.despine(left=True)
plt.grid(b=None, axis='x')
fig.savefig(join(analysis_path, 'mean_rankings_bar_chart.pdf'),
format='pdf', bbox_inches='tight')
plt.close()
def generate_main_results():
"""Generate the main results of the experiment."""
# Generate results
results = generate_results()
# Calculate results
mean_sem_scores = sort_tbl(
generate_mean_std_tbl(*calculate_mean_sem_scores(results)),
ovrs_order=OVERSAMPLERS_NAMES,
clfs_order=CLASSIFIERS_NAMES)
keys = mean_sem_scores[['Classifier', 'Metric']]
mean_sem_perc_diff_scores = []
for oversampler in ('SMOTE', 'K-MEANS SMOTE', 'SOMO', 'G-SMOTE'):
perc_diff_scores = sort_tbl(
generate_mean_std_tbl(*calculate_mean_sem_perc_diff_scores(
results, [oversampler, 'G-SOMO'])),
ovrs_order=OVERSAMPLERS_NAMES,
clfs_order=CLASSIFIERS_NAMES)
perc_diff_scores = perc_diff_scores.rename(columns={
'Difference': oversampler
}).drop(columns=['Classifier', 'Metric'])
mean_sem_perc_diff_scores.append(perc_diff_scores)
mean_sem_perc_diff_scores = pd.concat(
[keys, pd.concat(mean_sem_perc_diff_scores, axis=1)], axis=1)
mean_sem_ranking = sort_tbl(
generate_mean_std_tbl(*calculate_mean_sem_ranking(results)),
ovrs_order=OVERSAMPLERS_NAMES,
clfs_order=CLASSIFIERS_NAMES)
# Generate main results
main_results_names = ('mean_sem_scores', 'mean_sem_perc_diff_scores',
'mean_sem_ranking')
main_results = zip(
main_results_names,
(mean_sem_scores, mean_sem_perc_diff_scores, mean_sem_ranking))
return main_results