"""Tests for explainers.py — pure computation correctness.
Focused on compute_data_hash (D3-S2).
No model loading, no SHAP computation — fast and fully isolated.
"""
import pandas as pd
from xai_toolkit.explainers import compute_data_hash
class TestComputeDataHash:
"""Tests for compute_data_hash (D3-S2)."""
def _sample_df(self) -> pd.DataFrame:
"""Minimal DataFrame for deterministic testing."""
return pd.DataFrame(
{"feature_a": [1.0, 2.0, 3.0], "feature_b": [0.1, 0.2, 0.3]}
)
# --- Shape and format ---
def test_returns_64_char_hex_string(self):
"""SHA256 hex digest is always exactly 64 hex characters."""
h = compute_data_hash(self._sample_df())
assert isinstance(h, str)
assert len(h) == 64
def test_output_is_valid_hex(self):
"""Output contains only valid hex characters."""
h = compute_data_hash(self._sample_df())
assert all(c in "0123456789abcdef" for c in h)
# --- Determinism ---
def test_full_matrix_hash_is_deterministic(self):
"""Same DataFrame → same hash, always."""
df = self._sample_df()
assert compute_data_hash(df) == compute_data_hash(df)
def test_single_row_hash_is_deterministic(self):
"""Same row index → same hash, always."""
df = self._sample_df()
h1 = compute_data_hash(df, sample_index=0)
h2 = compute_data_hash(df, sample_index=0)
assert h1 == h2
def test_hash_is_stable_across_5_calls(self):
"""D3-S1: 5 consecutive calls produce the same hash."""
df = self._sample_df()
hashes = [compute_data_hash(df, sample_index=1) for _ in range(5)]
assert len(set(hashes)) == 1, (
f"Expected 1 unique hash across 5 calls, got {len(set(hashes))}"
)
# --- Sensitivity ---
def test_different_rows_produce_different_hashes(self):
"""Different rows with different values → different hashes."""
df = self._sample_df()
assert compute_data_hash(df, sample_index=0) != compute_data_hash(df, sample_index=1)
def test_full_matrix_differs_from_single_row(self):
"""Hashing the full matrix vs a single row must differ."""
df = self._sample_df()
assert compute_data_hash(df) != compute_data_hash(df, sample_index=0)
def test_tiny_value_change_changes_hash(self):
"""A single-digit change anywhere in the data changes the hash."""
df1 = pd.DataFrame({"a": [1.0], "b": [2.0]})
df2 = pd.DataFrame({"a": [1.0], "b": [2.0000001]})
assert compute_data_hash(df1) != compute_data_hash(df2)
