fix: correcting decomposition logic (#59)

Author: justushelo

panel/simdec_app.py

@@ -70,6 +70,7 @@ def load_data(text_fname):
         return pd.read_csv(io.BytesIO(raw))
     except Exception:
         pn.state.notifications.error(GENERIC_ERROR_MSG, duration=0)
+        return pd.read_csv(DEFAULT_STRESS_CSV)
 
 
 @pn.cache
@@ -183,11 +184,11 @@ def explained_variance_80(sensitivity_indices_table):
     si_values = df["Value"].tolist()[1:]
     input_names = df["Inputs"].tolist()[1:]
 
-    # Ensuring explained variance is at least 80% of the total
-    target = 0.8 * sum(si_values)
-    pos_80 = bisect.bisect_right(np.cumsum(si_values), target)
-
-    return input_names[: pos_80 + 1]
+    # Find the variables needed to reach 80% of explained variance
+    total = sum(si_values)
+    pos = bisect.bisect_left(np.cumsum(si_values), 0.8 * total)
+    n_vars = min(pos + 1, 4)
+    return input_names[:n_vars]
 
 
 @pn.cache

src/simdec/decomposition.py

@@ -138,7 +138,7 @@ def decomposition(
             n_var_dec = sensitivity_indices.size
 
         n_var_dec = max(1, n_var_dec)  # keep at least one variable
-        n_var_dec = min(5, n_var_dec)  # use at most 5 variables
+        n_var_dec = min(4, n_var_dec)  # use at most 4 variables
     else:
         n_var_dec = inputs.shape[1]
 

tests/test_decomposition.py

@@ -24,3 +24,52 @@ def test_decomposition():
     assert res.states == [2, 2, 2, 2]
     assert res.statistic.shape == (2, 2, 2, 2)
     npt.assert_allclose(res.bins.describe().T["mean"], res.statistic.flatten())
+
+
+def test_auto_ordering_single_dominant_variable():
+    fname = path_data / "stress.csv"
+    data = pd.read_csv(fname)
+    output_name, *v_names = list(data.columns)
+    inputs, output = data[v_names], data[output_name]
+
+    si = np.array([0.90, 0.05, 0.03, 0.02])
+    res = sd.decomposition(inputs=inputs, output=output, sensitivity_indices=si)
+    assert len(res.var_names) == 1
+
+
+def test_auto_ordering_two_variables_cross_threshold():
+    fname = path_data / "stress.csv"
+    data = pd.read_csv(fname)
+    output_name, *v_names = list(data.columns)
+    inputs, output = data[v_names], data[output_name]
+
+    # sum = 1.0, cumsum = [0.75, 0.81, ...] -> crosses 0.8 after 2nd variable
+    si = np.array([0.75, 0.06, 0.10, 0.09])
+    res = sd.decomposition(inputs=inputs, output=output, sensitivity_indices=si)
+    assert len(res.var_names) == 2
+
+
+def test_auto_ordering_cap_at_four():
+    """Even if more than 4 variables are needed to reach 0.8, cap at 4."""
+    fname = path_data / "stress.csv"
+    data = pd.read_csv(fname)
+    output_name, *v_names = list(data.columns)
+    inputs, output = data[v_names], data[output_name]
+
+    # sum = 1.0, each variable contributes equally -> need all 4 to reach 0.8
+    si = np.array([0.25, 0.25, 0.25, 0.25])
+    res = sd.decomposition(inputs=inputs, output=output, sensitivity_indices=si)
+    assert len(res.var_names) == 4
+
+
+def test_auto_ordering_si_not_summing_to_one():
+    """Threshold is relative to sum(si), not hardcoded 1.0."""
+    fname = path_data / "stress.csv"
+    data = pd.read_csv(fname)
+    output_name, *v_names = list(data.columns)
+    inputs, output = data[v_names], data[output_name]
+
+    # sum = 2.0, 0.8 * 2.0 = 1.6, cumsum = [1.8, ...] -> crosses after 1st variable
+    si = np.array([1.80, 0.10, 0.05, 0.05])
+    res = sd.decomposition(inputs=inputs, output=output, sensitivity_indices=si)
+    assert len(res.var_names) == 1