Fix: invalid value in divide

ncrf/_model.py

@@ -8,7 +8,7 @@
 from math import sqrt, log10
 from multiprocessing import current_process
 from operator import attrgetter
-from typing import Sequence, Union
+from typing import Iterator, Sequence, Union
 
 from eelbrain import fmtxt, UTS, NDVar
 import numpy as np
@@ -234,29 +234,38 @@ class RegressionData:
 
     Parameters
     ----------
-    tstart : float | list[float]
+    tstart
         Start of the TRF in seconds. Can define multiple tstarts for more than 1 predictor.
-    tstop : float | list[float]
+    tstop
         Stop of the TRF in seconds. Can define multiple tstops for more than 1 predictor.
-    nlevel : int
+    nlevel
         Decides the density of Gabor atoms. Bigger nlevel -> less dense basis.
         By default it is set to 1. ``nlevel > 2`` should be used with caution.
-    baseline: list | None
+    baseline
         Mean that will be subtracted from ``stim``.
-    scaling: list | None
+    scaling
         Scale by which ``stim`` was divided.
     """
     _n_predictor_variables = 1
     _prewhitened = None
 
-    def __init__(self, tstart, tstop, nlevel=1, baseline=None, scaling=None, stim_is_single=None, gaussian_fwhm=20.0):
+    def __init__(
+            self,
+            tstart: Union[float, list[float]],
+            tstop: Union[float, list[float]],
+            nlevel: int = 1,
+            baseline: list[float] = None,
+            scaling: list[float] = None,
+            stim_is_single: bool = None,
+            gaussian_fwhm: float = 20.0,
+    ):
         self.tstart = tstart if isinstance(tstart, collections.abc.Sequence) else [tstart]
         self.tstop = tstop if isinstance(tstop, collections.abc.Sequence) else [tstop]
         self.nlevel = nlevel
         self.s_baseline = baseline
         self.s_scaling = scaling
         self.s_normalization = []
-        self.meg = []
+        self.meg: list[np.ndarray] = []  # (sensor, time)
         self.covariates = []
         self.tstep = None
         self.filter_length = None
@@ -344,6 +353,11 @@ def add_data(self, meg, stim):
         m = max([basis.shape[0] for basis in self.basis])
         y = meg.get_data(('sensor', 'time'))
         y = y[:, m - 1:].astype(np.float64)
+        ch_var = np.var(y, axis=1)
+        zero_var = ch_var == 0
+        if zero_var.any():
+            flat_channels = self.sensor_dim.names[np.flatnonzero(zero_var)]
+            raise ValueError(f"{meg=}: data contains flat channels ({', '.join(flat_channels)})")
         self.meg.append(y / sqrt(y.shape[1]))  # Mind the normalization
 
         if self._norm_factor is None:
@@ -386,7 +400,7 @@ def _precompute(self):
             self._bE.append(np.dot(b, E))
             self._EtE.append(np.dot(E.T, E))
 
-    def __iter__(self):
+    def __iter__(self) -> Iterator[tuple[np.ndarray, np.ndarray]]:
         return zip(self.meg, self.covariates)
 
     def __len__(self):
@@ -1043,12 +1057,13 @@ def compute_explained_variance(self, data):
         logger.debug(f'{self.mu}: {1 - temp / len(data)}')
         return 1 - temp / len(data)
 
-    def _compute_voxelwise_explained_variance(self, data):
+    def _compute_voxelwise_explained_variance(self, data: RegressionData):
         """evaluates explained_variance
 
         Parameters
         ---------
-        data : REG_Data instance
+        data
+            Regression data.
 
         Returns
         -------