BUG: Raise early on non-finite values in PSD (Welch) and ICA.fit (Fixes #13364)

Author: zhijingz

doc/changes/devel/13364.bugfix.rst

@@ -0,0 +1,2 @@
+Fail early with a clear error when non-finite values (NaN/Inf) are present
+in PSD (Welch) and in ICA.fit, avoiding deep assertion failures (GH-13364).

mne/preprocessing/ica.py

@@ -891,6 +891,13 @@ def _pre_whiten(self, data):
 
     def _fit(self, data, fit_type):
         """Aux function."""
+        if not np.isfinite(data).all():
+            raise ValueError(
+                "Input data contains non-finite values (NaN/Inf). "
+                "Please clean your data (e.g., high-pass filter, interpolate or drop "
+                "contaminated segments) before calling ICA.fit()."
+            )
+
         random_state = check_random_state(self.random_state)
         n_channels, n_samples = data.shape
         self._compute_pre_whitener(data)

mne/preprocessing/tests/test_ica.py

@@ -1718,3 +1718,30 @@ def test_ica_ch_types(ch_type):
     for inst in [raw, epochs, evoked]:
         ica.apply(inst)
         ica.get_sources(inst)
+
+
+@pytest.mark.filterwarnings(
+    "ignore:The data has not been high-pass filtered.:RuntimeWarning"
+)
+@pytest.mark.filterwarnings(
+    "ignore:invalid value encountered in subtract:RuntimeWarning"
+)
+def test_ica_rejects_nonfinite():
+    """ICA.fit should fail early on NaN/Inf in the input data."""
+    info = create_info(["Fz", "Cz", "Pz", "Oz"], sfreq=100.0, ch_types="eeg")
+    rng = np.random.RandomState(1)
+    data = rng.randn(4, 1000)
+
+    # Case 1: NaN
+    raw = RawArray(data.copy(), info)
+    raw._data[0, 25] = np.nan
+    ica = ICA(n_components=2, random_state=0, method="fastica", max_iter="auto")
+    with pytest.raises(ValueError, match=r"Input data contains non[- ]?finite values"):
+        ica.fit(raw)
+
+    # Case 2: Inf
+    raw = RawArray(data.copy(), info)
+    raw._data[1, 50] = np.inf
+    ica = ICA(n_components=2, random_state=0, method="fastica", max_iter="auto")
+    with pytest.raises(ValueError, match=r"Input data contains non[- ]?finite values"):
+        ica.fit(raw)

mne/time_frequency/psd.py

@@ -200,6 +200,16 @@ def psd_array_welch(
     del freq_mask
     freqs = freqs[freq_sl]
 
+    # Hard error on Inf inside analyzed samples
+    step = max(n_per_seg - n_overlap, 1)
+    n_segments = 1 + (n_times - n_per_seg) // step if n_times >= n_per_seg else 0
+    analyzed_end = step * (n_segments - 1) + n_per_seg if n_segments > 0 else 0
+    if analyzed_end > 0 and np.isinf(x[..., :analyzed_end]).any():
+        raise ValueError(
+            "Input data contains non-finite values (Inf) in the analyzed time span. "
+            "Clean or drop bad segments before computing the PSD."
+        )
+
     # Parallelize across first N-1 dimensions
     logger.debug(
         f"Spectogram using {n_fft}-point FFT on {n_per_seg} samples with "
@@ -221,7 +231,12 @@ def psd_array_welch(
         good_mask = ~np.isnan(x)
         # NaNs originate from annot, so must match for all channels. Note that we CANNOT
         # use np.testing.assert_allclose() here; it is strict about shapes/broadcasting
-        assert np.allclose(good_mask, good_mask[[0]], equal_nan=True)
+        if not np.allclose(good_mask, good_mask[[0]], equal_nan=True):
+            raise ValueError(
+                "Input data contains NaN masks that are not aligned across channels; "
+                "make NaN spans consistent across channels or clean/drop bad segments."
+            )
+        # assert np.allclose(good_mask, good_mask[[0]], equal_nan=True)
         t_onsets, t_offsets = _mask_to_onsets_offsets(good_mask[0])
         x_splits = [x[..., t_ons:t_off] for t_ons, t_off in zip(t_onsets, t_offsets)]
         # weights reflect the number of samples used from each span. For spans longer

mne/time_frequency/tests/test_psd.py

@@ -226,3 +226,25 @@ def test_psd_array_welch_n_jobs():
     data = np.zeros((1, 2048))
     psd_array_welch(data, 1024, n_jobs=1)
     psd_array_welch(data, 1024, n_jobs=2)
+
+
+def test_psd_raises_on_inf_in_analyzed_window_array():
+    """psd_array_welch should fail if +Inf lies inside analyzed samples."""
+    n_samples, n_fft, n_overlap = 2048, 256, 128
+    rng = np.random.RandomState(0)
+    x = rng.randn(1, n_samples)
+    # Put +Inf inside the series; this falls within Welch windows
+    x[0, 800] = np.inf
+    with pytest.raises(ValueError, match="non[- ]?finite|NaN|Inf"):
+        psd_array_welch(x, float(n_fft), n_fft=n_fft, n_overlap=n_overlap)
+
+
+def test_psd_raises_on_misaligned_nan_across_channels():
+    """If NaNs are present but masks are NOT aligned across channels, raise."""
+    n_samples, n_fft, n_overlap = 2048, 256, 128
+    rng = np.random.RandomState(42)
+    x = rng.randn(2, n_samples)
+    # NaN only in ch0; ch1 has no NaN => masks not aligned -> should raise
+    x[0, 500] = np.nan
+    with pytest.raises(ValueError, match="aligned|not aligned|non[- ]?finite|NaN|Inf"):
+        psd_array_welch(x, float(n_fft), n_fft=n_fft, n_overlap=n_overlap)