commit-message-editing-visualization / dataset_statistics.py
Petr Tsvetkov
Fix the synthetic data generation pipeline
347f566
raw
history blame
2.48 kB
import pickle
import Levenshtein
import numpy as np
import pandas as pd
import plotly.figure_factory as ff
from scipy.stats import stats
import config
def get_statistics_for_sample(start_msg, end_msg, row=None):
edit_ops = Levenshtein.editops(start_msg, end_msg)
n_deletes = sum([1 if op == 'delete' else 0 for op, _, _ in edit_ops])
n_inserts = sum([1 if op == 'insert' else 0 for op, _, _ in edit_ops])
n_replaces = sum([1 if op == 'replace' else 0 for op, _, _ in edit_ops])
n_changes = n_deletes + n_inserts + n_replaces
n_deletes += n_replaces
n_inserts += n_replaces
return {
"deletions": n_deletes,
"insertions": n_inserts,
"changes": n_changes,
"deletions_norm": n_deletes / len(start_msg),
"insertions_norm": n_inserts / len(end_msg),
"changes_norm": n_changes / len(end_msg),
"lendiff": abs(len(start_msg) - len(end_msg)),
"editdist": row["editdist_related"] if row is not None else Levenshtein.distance(start_msg, end_msg),
}
def get_statistics_for_row(row):
start_msg = row["commit_msg_start"]
end_msg = row["commit_msg_end"]
return get_statistics_for_sample(start_msg, end_msg, row=row)
def get_statistics_for_df(df: pd.DataFrame):
stats = [get_statistics_for_row(row) for _, row in
df.iterrows()]
assert len(stats) > 0
return {stat_name: np.asarray([e[stat_name] for e in stats]) for stat_name in stats[0]}
def build_plotly_chart(stat_golden, stat_e2s, stat_s2e, stat_e2s_s2e, stat_name):
hist_data = [stat_golden, stat_e2s, stat_s2e, stat_e2s_s2e,
np.concatenate((stat_e2s, stat_s2e, stat_e2s_s2e), axis=0)]
group_labels = ['Golden', 'e2s', 's2e', 'e2s+s2e', 'Synthetic']
fig = ff.create_distplot(hist_data, group_labels,
bin_size=.05, show_rug=False, show_hist=False)
fig.update_layout(title_text=stat_name)
with open(config.OUTPUT_CHARTS_DIR / f"{stat_name}_data.pkl", "wb") as f:
pickle.dump(hist_data, f)
return fig
def t_test(group_stats, main_group="manual"):
results = {}
for group in group_stats:
results[group] = {}
for stat in group_stats[group]:
a = group_stats[main_group][stat]
b = group_stats[group][stat]
p = stats.ttest_ind(a, b, equal_var=False, random_state=config.RANDOM_STATE).pvalue
results[group][stat] = p
return results