DOC: Better documentation of realign_raw (mne-tools#12135)

mne/epochs.py

@@ -4062,12 +4062,13 @@ def _concatenate_epochs(
     event_id = deepcopy(out.event_id)
     selection = out.selection
     # offset is the last epoch + tmax + 10 second
-    shift = int((10 + tmax) * out.info["sfreq"])
+    shift = np.int64((10 + tmax) * out.info["sfreq"])
     # Allow reading empty epochs (ToDo: Maybe not anymore in the future)
     if out._allow_empty:
         events_offset = 0
     else:
         events_offset = int(np.max(events[0][:, 0])) + shift
+    events_offset = np.int64(events_offset)
     events_overflow = False
     warned = False
     for ii, epochs in enumerate(epochs_list[1:], 1):

mne/io/bti/read.py

@@ -48,12 +48,12 @@ def read_int8(fid):
 
 def read_uint16(fid):
     """Read unsigned 16bit integer from bti file."""
-    return _unpack_simple(fid, ">u2", np.uint16)
+    return _unpack_simple(fid, ">u2", np.uint32)
 
 
 def read_int16(fid):
     """Read 16bit integer from bti file."""
-    return _unpack_simple(fid, ">i2", np.int16)
+    return _unpack_simple(fid, ">i2", np.int32)
 
 
 def read_uint32(fid):
@@ -88,7 +88,13 @@ def read_double(fid):
 
 def read_int16_matrix(fid, rows, cols):
     """Read 16bit integer matrix from bti file."""
-    return _unpack_matrix(fid, rows, cols, dtype=">i2", out_dtype=np.int16)
+    return _unpack_matrix(
+        fid,
+        rows,
+        cols,
+        dtype=">i2",
+        out_dtype=np.int32,
+    )
 
 
 def read_float_matrix(fid, rows, cols):

mne/io/ctf/res4.py

@@ -43,7 +43,7 @@ def _read_ustring(fid, n_bytes):
 
 def _read_int2(fid):
     """Read int from short."""
-    return np.fromfile(fid, ">i2", 1)[0]
+    return _auto_cast(np.fromfile(fid, ">i2", 1)[0])
 
 
 def _read_int(fid):
@@ -208,6 +208,9 @@ def _read_res4(dsdir):
             coil["area"] *= 1e-4
         # convert to dict
         chs = [dict(zip(chs.dtype.names, x)) for x in chs]
+        for ch in chs:
+            for key, val in ch.items():
+                ch[key] = _auto_cast(val)
         res["chs"] = chs
         for k in range(res["nchan"]):
             res["chs"][k]["ch_name"] = res["ch_names"][k]
@@ -216,3 +219,15 @@ def _read_res4(dsdir):
         _read_comp_coeff(fid, res)
     logger.info("    res4 data read.")
     return res
+
+
+def _auto_cast(x):
+    # Upcast scalars
+    if isinstance(x, np.ScalarType):
+        if x.dtype.kind == "i":
+            if x.dtype != np.int64:
+                x = x.astype(np.int64)
+        elif x.dtype.kind == "f":
+            if x.dtype != np.float64:
+                x = x.astype(np.float64)
+    return x

mne/io/edf/edf.py

@@ -1106,7 +1106,7 @@ def _read_gdf_header(fname, exclude, include=None):
                     "Header information is incorrect for record length. "
                     "Default record length set to 1."
                 )
-            nchan = np.fromfile(fid, UINT32, 1)[0]
+            nchan = int(np.fromfile(fid, UINT32, 1)[0])
             channels = list(range(nchan))
             ch_names = [_edf_str(fid.read(16)).strip() for ch in channels]
             exclude = _find_exclude_idx(ch_names, exclude, include)
@@ -1177,7 +1177,7 @@ def _read_gdf_header(fname, exclude, include=None):
             fid.seek(etp)
             etmode = np.fromfile(fid, UINT8, 1)[0]
             if etmode in (1, 3):
-                sr = np.fromfile(fid, UINT8, 3)
+                sr = np.fromfile(fid, UINT8, 3).astype(np.uint32)
                 event_sr = sr[0]
                 for i in range(1, len(sr)):
                     event_sr = event_sr + sr[i] * 2 ** (i * 8)
@@ -1297,7 +1297,7 @@ def _read_gdf_header(fname, exclude, include=None):
                     "Default record length set to 1."
                 )
 
-            nchan = np.fromfile(fid, UINT16, 1)[0]
+            nchan = int(np.fromfile(fid, UINT16, 1)[0])
             fid.seek(2, 1)  # 2bytes reserved
 
             # Channels (variable header)

mne/io/egi/egi.py

@@ -29,7 +29,7 @@ def _read_header(fid):
         )
 
     def my_fread(*x, **y):
-        return np.fromfile(*x, **y)[0]
+        return int(np.fromfile(*x, **y)[0])
 
     info = dict(
         version=version,
@@ -57,8 +57,8 @@ def my_fread(*x, **y):
             dict(
                 n_categories=0,
                 n_segments=1,
-                n_samples=np.fromfile(fid, ">i4", 1)[0],
-                n_events=np.fromfile(fid, ">i2", 1)[0],
+                n_samples=int(np.fromfile(fid, ">i4", 1)[0]),
+                n_events=int(np.fromfile(fid, ">i2", 1)[0]),
                 event_codes=[],
                 category_names=[],
                 category_lengths=[],

mne/io/egi/egimff.py

@@ -79,7 +79,7 @@ def _read_mff_header(filepath):
         # by what we need to (e.g., a sample rate of 500 means we can multiply
         # by 1 and divide by 2 rather than multiplying by 500 and dividing by
         # 1000)
-        numerator = signal_blocks["sfreq"]
+        numerator = int(signal_blocks["sfreq"])
         denominator = 1000
         this_gcd = math.gcd(numerator, denominator)
         numerator = numerator // this_gcd

mne/io/nihon/nihon.py

@@ -207,7 +207,7 @@ def _read_nihon_header(fname):
                 t_datablock["address"] = t_data_address
 
                 fid.seek(t_data_address + 0x26)
-                t_n_channels = np.fromfile(fid, np.uint8, 1)[0]
+                t_n_channels = np.fromfile(fid, np.uint8, 1)[0].astype(np.int64)
                 t_datablock["n_channels"] = t_n_channels
 
                 t_channels = []
@@ -219,14 +219,14 @@ def _read_nihon_header(fname):
                 t_datablock["channels"] = t_channels
 
                 fid.seek(t_data_address + 0x1C)
-                t_record_duration = np.fromfile(fid, np.uint32, 1)[0]
+                t_record_duration = np.fromfile(fid, np.uint32, 1)[0].astype(np.int64)
                 t_datablock["duration"] = t_record_duration
 
                 fid.seek(t_data_address + 0x1A)
                 sfreq = np.fromfile(fid, np.uint16, 1)[0] & 0x3FFF
-                t_datablock["sfreq"] = sfreq
+                t_datablock["sfreq"] = sfreq.astype(np.int64)
 
-                t_datablock["n_samples"] = int(t_record_duration * sfreq / 10)
+                t_datablock["n_samples"] = np.int64(t_record_duration * sfreq // 10)
                 t_controlblock["datablocks"].append(t_datablock)
             controlblocks.append(t_controlblock)
         header["controlblocks"] = controlblocks

mne/io/nsx/nsx.py

@@ -365,7 +365,7 @@ def _get_hdr_info(fname, stim_channel=True, eog=None, misc=None):
 
     stim_channel_idxs, _ = _check_stim_channel(stim_channel, ch_names)
 
-    nchan = nsx_info["channel_count"]
+    nchan = int(nsx_info["channel_count"])
     logger.info("Setting channel info structure...")
     chs = list()
     pick_mask = np.ones(len(ch_names))

mne/preprocessing/realign.py

@@ -28,8 +28,9 @@ def realign_raw(raw, other, t_raw, t_other, verbose=None):
         The second raw instance. It will be resampled to match ``raw``.
     t_raw : array-like, shape (n_events,)
         The times of shared events in ``raw`` relative to ``raw.times[0]`` (0).
-        Typically these could be events on some TTL channel like
-        ``find_events(raw)[:, 0] - raw.first_samp``.
+        Typically these could be events on some TTL channel such as::
+
+            find_events(raw)[:, 0] / raw.info["sfreq"] - raw.first_time
     t_other : array-like, shape (n_events,)
         The times of shared events in ``other`` relative to ``other.times[0]``.
     %(verbose)s
@@ -92,11 +93,11 @@ def realign_raw(raw, other, t_raw, t_other, verbose=None):
         logger.info(f"Cropping {zero_ord:0.3f} s from the start of raw")
         raw.crop(zero_ord, None)
         t_raw -= zero_ord
-    else:  # need to crop start of other to match raw
-        t_crop = zero_ord / first_ord
+    elif zero_ord < 0:  # need to crop start of other to match raw
+        t_crop = -zero_ord / first_ord
         logger.info(f"Cropping {t_crop:0.3f} s from the start of other")
-        other.crop(-t_crop, None)
-        t_other += t_crop
+        other.crop(t_crop, None)
+        t_other -= t_crop
 
     # 3. Resample data using the first-order term
     nan_ch_names = [

mne/report/tests/test_report.py

@@ -877,6 +877,8 @@ def test_survive_pickle(tmp_path):
 def test_manual_report_2d(tmp_path, invisible_fig):
     """Simulate user manually creating report by adding one file at a time."""
     pytest.importorskip("sklearn")
+    pytest.importorskip("pandas")
+
     from sklearn.exceptions import ConvergenceWarning
 
     r = Report(title="My Report")

mne/surface.py

@@ -772,7 +772,7 @@ def _call_old(self, rr, n_jobs):
 
 def _fread3(fobj):
     """Read 3 bytes and adjust."""
-    b1, b2, b3 = np.fromfile(fobj, ">u1", 3)
+    b1, b2, b3 = np.fromfile(fobj, ">u1", 3).astype(np.int64)
     return (b1 << 16) + (b2 << 8) + b3
 
 

tools/azure_dependencies.sh

@@ -9,7 +9,8 @@ elif [ "${TEST_MODE}" == "pip-pre" ]; then
 	python -m pip install $STD_ARGS pip setuptools wheel packaging setuptools_scm
 	python -m pip install $STD_ARGS --only-binary ":all:" --extra-index-url "https://www.riverbankcomputing.com/pypi/simple" PyQt6 PyQt6-sip PyQt6-Qt6
 	echo "Numpy etc."
-	python -m pip install $STD_ARGS --only-binary ":all:" --extra-index-url "https://pypi.anaconda.org/scientific-python-nightly-wheels/simple" "numpy>=2.0.0.dev0" "scipy>=1.12.0.dev0" statsmodels pandas scikit-learn matplotlib
+	# As of 2023/10/25 no pandas (or statsmodels) because they pin to NumPy < 2
+	python -m pip install $STD_ARGS --only-binary ":all:" --extra-index-url "https://pypi.anaconda.org/scientific-python-nightly-wheels/simple" "numpy>=2.0.0.dev0" "scipy>=1.12.0.dev0" scikit-learn matplotlib
 	echo "dipy"
 	python -m pip install $STD_ARGS --only-binary ":all:" --extra-index-url "https://pypi.anaconda.org/scipy-wheels-nightly/simple" dipy
 	echo "h5py"

tools/github_actions_dependencies.sh

@@ -1,5 +1,7 @@
 #!/bin/bash -ef
 
+set -o pipefail
+
 STD_ARGS="--progress-bar off --upgrade"
 if [ ! -z "$CONDA_ENV" ]; then
 	echo "Uninstalling MNE for CONDA_ENV=${CONDA_ENV}"
@@ -18,7 +20,8 @@ else
 	echo "PyQt6"
 	pip install $STD_ARGS --only-binary ":all:" --default-timeout=60 --extra-index-url https://www.riverbankcomputing.com/pypi/simple PyQt6
 	echo "NumPy/SciPy/pandas etc."
-	pip install $STD_ARGS --only-binary ":all:" --default-timeout=60 --extra-index-url "https://pypi.anaconda.org/scientific-python-nightly-wheels/simple" "numpy>=2.0.0.dev0" scipy scikit-learn pandas matplotlib pillow statsmodels
+	# As of 2023/10/25 no pandas (or statsmodels, nilearn) because they pin to NumPy < 2
+	pip install $STD_ARGS --only-binary ":all:" --default-timeout=60 --extra-index-url "https://pypi.anaconda.org/scientific-python-nightly-wheels/simple" "numpy>=2.0.0.dev0" scipy scikit-learn matplotlib pillow
 	echo "dipy"
 	pip install $STD_ARGS --only-binary ":all:" --default-timeout=60 --extra-index-url "https://pypi.anaconda.org/scipy-wheels-nightly/simple" dipy
 	echo "H5py"
@@ -27,7 +30,8 @@ else
 	pip install $STD_ARGS --only-binary ":all:" --extra-index-url "https://test.pypi.org/simple" openmeeg
 	# No Numba because it forces an old NumPy version
 	echo "nilearn and openmeeg"
-	pip install $STD_ARGS git+https://github.com/nilearn/nilearn
+	# pip install $STD_ARGS git+https://github.com/nilearn/nilearn
+	pip install $STD_ARGS openmeeg
 	echo "VTK"
 	pip install $STD_ARGS --only-binary ":all:" --extra-index-url "https://wheels.vtk.org" vtk
 	python -c "import vtk"
@@ -45,9 +49,12 @@ else
 	pip install $STD_ARGS git+https://github.com/joblib/joblib@master
 	echo "EDFlib-Python"
 	pip install $STD_ARGS git+https://gitlab.com/Teuniz/EDFlib-Python@master
+	# Until Pandas is fixed, make sure we didn't install it
+	! python -c "import pandas"
 fi
 echo ""
 
+
 # for compat_minimal and compat_old, we don't want to --upgrade
 if [ ! -z "$CONDA_DEPENDENCIES" ]; then
 	echo "Installing dependencies for conda"