def test_compute_statistics_on_empty_set():
result = compute_statistics([])
assert result["count"] == 0
assert all(
np.isnan(value) for key, value in result.items() if key != "count")
# we see the same keys as in those cases where we can evaluate something
dummy_result_non_empty = compute_statistics([1, 2])
assert set(dummy_result_non_empty.keys()) == set(result.keys())
def test_process_results_per_session(seed: int):
rng = np.random.default_rng(seed=seed)
(
per_session_results,
preceding_user_turn_numbers_used_per_marker,
) = _generate_random_examples(num_markers=3, rng=rng)
markers = sorted(preceding_user_turn_numbers_used_per_marker.keys())
num_sessions = len(per_session_results)
stats = MarkerStatistics()
sender_ids = []
session_indices = []
for session_idx, results in enumerate(per_session_results):
sender_id = str(rng.choice(100))
session_idx = int(rng.choice(100))
stats.process(
session_idx=session_idx,
sender_id=sender_id,
meta_data_on_relevant_events_per_marker=results,
)
sender_ids.append(sender_id)
session_indices.append(session_idx)
assert stats.num_sessions == len(per_session_results)
for marker in markers:
for idx in range(num_sessions):
expected_stats = compute_statistics(
preceding_user_turn_numbers_used_per_marker[marker][idx])
for stat_name, stat_value in expected_stats.items():
assert pytest.approx(
stats.session_results[marker][stat_name][idx], stat_value)
for idx in range(num_sessions):
assert stats.session_identifier[idx] == (sender_ids[idx],
session_indices[idx])
def test_compute_statistics_simple_check():
stats = compute_statistics([1, 2, 9, 0])
assert stats["count"] == 4
assert stats["min"] == 0
assert stats["max"] == 9
assert stats["mean"] == 3
assert stats[
"median"] == 1.5 # this is no bug, it is a convention numpy follows
def test_overall_statistics_to_csv(tmp_path: Path, seed: int):
rng = np.random.default_rng(seed=seed)
(
per_session_results,
preceding_user_turn_numbers_used_per_marker,
) = _generate_random_examples(num_markers=3,
rng=rng,
num_sessions_min=10,
num_sessions_max=20)
markers = sorted(preceding_user_turn_numbers_used_per_marker.keys())
num_sessions = len(per_session_results)
stats = MarkerStatistics()
for session_idx, results in enumerate(per_session_results):
stats.process(
session_idx=session_idx,
sender_id=str(rng.choice(100)),
meta_data_on_relevant_events_per_marker=results,
)
tmp_file = tmp_path / "test.csv"
stats.overall_statistic_to_csv(path=tmp_file)
with tmp_file.open(mode="r") as f:
reader = csv.DictReader(f)
rows = [row for row in reader]
assert rows[0] == {
"sender_id": "all",
"session_idx": "nan",
"marker": "-",
"statistic": "total_number_of_sessions",
"value": str(num_sessions),
}
num_digits = 3
row_idx = 1
for marker_name in markers:
assert rows[row_idx] == {
"sender_id": "all",
"session_idx": "nan",
"marker": marker_name,
"statistic":
"number_of_sessions_where_marker_applied_at_least_once",
"value": str(stats.count_if_applied_at_least_once[marker_name]),
}
row_idx += 1
assert rows[row_idx] == {
"sender_id":
"all",
"session_idx":
"nan",
"marker":
marker_name,
"statistic":
"percentage_of_sessions_where_marker_applied_at_least_once",
"value":
str(
round(
stats.count_if_applied_at_least_once[marker_name] /
num_sessions * 100,
num_digits,
)),
}
row_idx += 1
for marker_name in markers:
statistics = compute_statistics(
stats.num_preceding_user_turns_collected[marker_name])
for stat_name, stat_value in statistics.items():
assert rows[row_idx] == {
"sender_id": "all",
"session_idx": "nan",
"marker": marker_name,
"statistic":
MarkerStatistics._add_num_user_turns_str_to(stat_name),
"value": str(round(stat_value, num_digits)),
}
row_idx += 1