Merge pull request #1797 from Kkuntal990/BrainVisionIO-vectorized-fix

Add support for VECTORIZED orientation in BrainVisionRawIO

Author: samuelgarcia

neo/rawio/baserawio.py

@@ -1577,7 +1577,9 @@ def __init__(self, *arg, **kwargs):
     def _get_signal_size(self, block_index, seg_index, stream_index):
         buffer_id = self.header["signal_streams"][stream_index]["buffer_id"]
         buffer_desc = self.get_analogsignal_buffer_description(block_index, seg_index, buffer_id)
-        # some hdf5 revert teh buffer
+        # time_axis indicates which dimension is time:
+        # time_axis=0: shape is (time, channels)
+        # time_axis=1: shape is (channels, time)
         time_axis = buffer_desc.get("time_axis", 0)
         return buffer_desc["shape"][time_axis]
 
@@ -1598,34 +1600,67 @@ def _get_analogsignal_chunk(
 
         buffer_desc = self.get_analogsignal_buffer_description(block_index, seg_index, buffer_id)
 
+        # Get time_axis to determine which dimension is time
+        time_axis = buffer_desc.get("time_axis", 0)
+
         i_start = i_start or 0
-        i_stop = i_stop or buffer_desc["shape"][0]
+        i_stop = i_stop or buffer_desc["shape"][time_axis]
 
         if buffer_desc["type"] == "raw":
 
-            # open files on demand and keep reference to opened file
-            if not hasattr(self, "_memmap_analogsignal_buffers"):
-                self._memmap_analogsignal_buffers = {}
-            if block_index not in self._memmap_analogsignal_buffers:
-                self._memmap_analogsignal_buffers[block_index] = {}
-            if seg_index not in self._memmap_analogsignal_buffers[block_index]:
-                self._memmap_analogsignal_buffers[block_index][seg_index] = {}
-            if buffer_id not in self._memmap_analogsignal_buffers[block_index][seg_index]:
-                fid = open(buffer_desc["file_path"], mode="rb")
-                self._memmap_analogsignal_buffers[block_index][seg_index][buffer_id] = fid
-            else:
-                fid = self._memmap_analogsignal_buffers[block_index][seg_index][buffer_id]
+            if time_axis == 0:
+                # MULTIPLEXED: time_axis=0 means (time, channels) layout
+                # open files on demand and keep reference to opened file
+                if not hasattr(self, "_memmap_analogsignal_buffers"):
+                    self._memmap_analogsignal_buffers = {}
+                if block_index not in self._memmap_analogsignal_buffers:
+                    self._memmap_analogsignal_buffers[block_index] = {}
+                if seg_index not in self._memmap_analogsignal_buffers[block_index]:
+                    self._memmap_analogsignal_buffers[block_index][seg_index] = {}
+                if buffer_id not in self._memmap_analogsignal_buffers[block_index][seg_index]:
+                    fid = open(buffer_desc["file_path"], mode="rb")
+                    self._memmap_analogsignal_buffers[block_index][seg_index][buffer_id] = fid
+                else:
+                    fid = self._memmap_analogsignal_buffers[block_index][seg_index][buffer_id]
+
+                num_channels = buffer_desc["shape"][1]
+
+                raw_sigs = get_memmap_chunk_from_opened_file(
+                    fid,
+                    num_channels,
+                    i_start,
+                    i_stop,
+                    np.dtype(buffer_desc["dtype"]),
+                    file_offset=buffer_desc["file_offset"],
+                )
 
-            num_channels = buffer_desc["shape"][1]
+            elif time_axis == 1:
+                # VECTORIZED: time_axis=1 means shape is (channels, time)
+                # Data is stored as [all_samples_ch1, all_samples_ch2, ...]
+                dtype = np.dtype(buffer_desc["dtype"])
+                num_channels = buffer_desc["shape"][0]  # shape is (channels, time)
+                num_samples = i_stop - i_start
+                total_samples_per_channel = buffer_desc["shape"][1]  # shape is (channels, time)
+
+                # Determine which channels to read
+                if channel_indexes is None:
+                    chan_indices = np.arange(num_channels)
+                else:
+                    chan_indices = np.asarray(channel_indexes)
+
+                raw_sigs = np.empty((num_samples, len(chan_indices)), dtype=dtype)
+
+                for i, chan_idx in enumerate(chan_indices):
+                    offset = buffer_desc["file_offset"] + chan_idx * total_samples_per_channel * dtype.itemsize
+                    channel_data = np.memmap(buffer_desc["file_path"], dtype=dtype, mode='r',
+                                            offset=offset, shape=(total_samples_per_channel,))
+                    raw_sigs[:, i] = channel_data[i_start:i_stop]
+
+                # Channel slicing already done above, so skip later channel_indexes slicing
+                channel_indexes = None
 
-            raw_sigs = get_memmap_chunk_from_opened_file(
-                fid,
-                num_channels,
-                i_start,
-                i_stop,
-                np.dtype(buffer_desc["dtype"]),
-                file_offset=buffer_desc["file_offset"],
-            )
+            else:
+                raise ValueError(f"time_axis must be 0 or 1, got {time_axis}")
 
         elif buffer_desc["type"] == "hdf5":
 

neo/rawio/brainvisionrawio.py

@@ -62,9 +62,12 @@ def _parse_header(self):
             raise NeoReadWriteError(
                 f"Only `BINARY` format has been implemented. Current Data Format is {vhdr_header['Common Infos']['DataFormat']}"
             )
-        if vhdr_header["Common Infos"]["DataOrientation"] != "MULTIPLEXED":
+
+        # Store the data orientation for later use in reading
+        self._data_orientation = vhdr_header["Common Infos"]["DataOrientation"]
+        if self._data_orientation not in ("MULTIPLEXED", "VECTORIZED"):
             raise NeoReadWriteError(
-                f"Only `MULTIPLEXED` is implemented. Current Orientation is {vhdr_header['Common Infos']['DataOrientation']}"
+                f"Data orientation must be either `MULTIPLEXED` or `VECTORIZED`. Current Orientation is {self._data_orientation}"
             )
 
         nb_channel = int(vhdr_header["Common Infos"]["NumberOfChannels"])
@@ -87,14 +90,27 @@ def _parse_header(self):
         buffer_id = "0"
         self._buffer_descriptions = {0: {0: {}}}
         self._stream_buffer_slice = {}
-        shape = get_memmap_shape(binary_filename, sig_dtype, num_channels=nb_channel, offset=0)
+
+        # time_axis indicates data layout: 0 for MULTIPLEXED (time, channels), 1 for VECTORIZED (channels, time)
+        time_axis = 0 if self._data_orientation == "MULTIPLEXED" else 1
+
+        # Get shape - always returns (num_samples, num_channels)
+        temp_shape = get_memmap_shape(binary_filename, sig_dtype, num_channels=nb_channel, offset=0)
+
+        # For consistency with HDF5 pattern: when time_axis=1, shape should be (channels, time)
+        if time_axis == 1:
+            shape = (temp_shape[1], temp_shape[0])  # (num_channels, num_samples)
+        else:
+            shape = temp_shape  # (num_samples, num_channels)
+
         self._buffer_descriptions[0][0][buffer_id] = {
             "type": "raw",
             "file_path": binary_filename,
             "dtype": str(sig_dtype),
             "order": "C",
             "file_offset": 0,
             "shape": shape,
+            "time_axis": time_axis,
         }
         self._stream_buffer_slice[stream_id] = None