Adding/Removing Containers from an NWB File

This tutorial explains how to add and remove containers from an existing NWB file and either write the data back to the same file or export the data to a new file.

Adding objects to an NWB file in read/write mode

PyNWB supports adding container objects to an existing NWB file - that is, reading data from an NWB file, adding a container object, such as a new TimeSeries object, and writing the modified NWBFile back to the same file path on disk. To do so:

1. open the file with an NWBHDF5IO object in read/write mode (mode='r+' or mode='a')

2. read the NWBFile

3. add container objects to the NWBFile object

4. write the modified NWBFile using the same NWBHDF5IO object

For example:

import datetime

import numpy as np

from pynwb import NWBHDF5IO, NWBFile, TimeSeries

# first, write a test NWB file
nwbfile = NWBFile(
    session_description="demonstrate adding to an NWB file",
    identifier="NWB123",
    session_start_time=datetime.datetime.now(datetime.timezone.utc),
)

filename = "nwbfile.nwb"
with NWBHDF5IO(filename, "w") as io:
    io.write(nwbfile)

# open the NWB file in r+ mode
with NWBHDF5IO(filename, "r+") as io:
    read_nwbfile = io.read()

    # create a TimeSeries and add it to the file under the acquisition group
    data = list(range(100, 200, 10))
    timestamps = np.arange(10, dtype=float)
    test_ts = TimeSeries(
        name="test_timeseries", data=data, unit="m", timestamps=timestamps
    )
    read_nwbfile.add_acquisition(test_ts)

    # write the modified NWB file
    io.write(read_nwbfile)

# confirm the file contains the new TimeSeries in acquisition
with NWBHDF5IO(filename, "r") as io:
    read_nwbfile = io.read()
    print(read_nwbfile)

Note

You cannot remove objects from an NWB file using the above method.

Modifying an NWB file in this way has limitations. The destination file path must be the same as the source file path, and it is not possible to remove objects from an NWB file. You can use the NWBHDF5IO.export method, detailed below, to modify an NWB file in these ways.

Exporting a written NWB file to a new file path

Use the NWBHDF5IO.export method to read data from an existing NWB file, modify the data, and write the modified data to a new file path. Modifications to the data can be additions or removals of objects, such as TimeSeries objects. This is especially useful if you have raw data and processed data in the same NWB file and you want to create a new NWB file with all the contents of the original file except for the raw data for sharing with collaborators.

To remove existing containers, use the pop method on any LabelledDict object, such as NWBFile.acquisition, NWBFile.processing, NWBFile.analysis, NWBFile.processing, NWBFile.scratch, NWBFile.devices, NWBFile.stimulus, NWBFile.stimulus_template, NWBFile.electrode_groups, NWBFile.imaging_planes, NWBFile.icephys_electrodes, NWBFile.ogen_sites, NWBFile.lab_meta_data, and ProcessingModule objects.

For example:

# first, create a test NWB file with a TimeSeries in the acquisition group
nwbfile = NWBFile(
    session_description="demonstrate export of an NWB file",
    identifier="NWB123",
    session_start_time=datetime.datetime.now(datetime.timezone.utc),
)
data1 = list(range(100, 200, 10))
timestamps1 = np.arange(10, dtype=float)
test_ts1 = TimeSeries(
    name="test_timeseries1", data=data1, unit="m", timestamps=timestamps1
)
nwbfile.add_acquisition(test_ts1)

# then, create a processing module for processed behavioral data
nwbfile.create_processing_module(
    name="behavior", description="processed behavioral data"
)
data2 = list(range(100, 200, 10))
timestamps2 = np.arange(10, dtype=float)
test_ts2 = TimeSeries(
    name="test_timeseries2", data=data2, unit="m", timestamps=timestamps2
)
nwbfile.processing["behavior"].add(test_ts2)

# write these objects to an NWB file
filename = "nwbfile.nwb"
with NWBHDF5IO(filename, "w") as io:
    io.write(nwbfile)

# read the written file
export_filename = "exported_nwbfile.nwb"
with NWBHDF5IO(filename, mode="r") as read_io:
    read_nwbfile = read_io.read()

    # add a new TimeSeries to the behavior processing module
    data3 = list(range(100, 200, 10))
    timestamps3 = np.arange(10, dtype=float)
    test_ts3 = TimeSeries(
        name="test_timeseries3", data=data3, unit="m", timestamps=timestamps3
    )
    read_nwbfile.processing["behavior"].add(test_ts3)

    # use the pop method to remove the original TimeSeries from the acquisition group
    read_nwbfile.acquisition.pop("test_timeseries1")

    # use the pop method to remove a TimeSeries from a processing module
    read_nwbfile.processing["behavior"].data_interfaces.pop("test_timeseries2")

    # call the export method to write the modified NWBFile instance to a new file path.
    # the original file is not modified
    with NWBHDF5IO(export_filename, mode="w") as export_io:
        export_io.export(src_io=read_io, nwbfile=read_nwbfile)

# confirm the exported file does not contain TimeSeries with names 'test_timeseries1' or 'test_timeseries2'
# but does contain a new TimeSeries in processing['behavior'] with name 'test_timeseries3'
with NWBHDF5IO(export_filename, "r") as io:
    read_nwbfile = io.read()
    print(read_nwbfile)
    print(read_nwbfile.processing["behavior"])

Note

TimeIntervals objects, such as NWBFile.epochs, NWBFile.trials, NWBFile.invalid_times, and custom TimeIntervals objects cannot be removed (popped) from NWBFile.intervals.

Warning

Removing an object from an NWBFile may break links and references within the file and across files. This is analogous to having shortcuts/aliases to a file on your filesystem and then deleting the file. Extra caution should be taken when removing heavily referenced items such as Device objects, ElectrodeGroup objects, the electrodes table, and the PlaneSegmentation table.

Exporting with new object IDs

When exporting a read NWB file to a new file path, the object IDs within the original NWB file will be copied to the new file. To make the exported NWB file contain a new set of object IDs, call generate_new_id on your NWBFile object. This will generate a new object ID for the NWBFile object and all of the objects within the NWB file.

export_filename = "exported_nwbfile.nwb"
with NWBHDF5IO(filename, mode="r") as read_io:
    read_nwbfile = read_io.read()
    read_nwbfile.generate_new_id()

    with NWBHDF5IO(export_filename, mode="w") as export_io:
        export_io.export(src_io=read_io, nwbfile=read_nwbfile)

Upgrading a legacy Device.model string to a DeviceModel

NWB Schema 2.9 changed Device.model from a string to a link to a DeviceModel. When you read a file written with an older version of the schema, a string Device.model is upgraded to a DeviceModel whose name is the original string, so that it conforms to the latest schema. PyNWB issues a warning when this upgrade happens.

NWB object names cannot contain / or :. If the legacy model string contained either character (e.g., "MFC_200/250-0.66_40mm_MF2.5:FLT"), the upgraded DeviceModel is read-only: building it for write or export raises an error, because the invalid name would otherwise be interpreted as nested HDF5 groups and corrupt the file.

The example below first synthesizes a legacy file by writing a Device and then adding the string model attribute directly with h5py (PyNWB cannot write a string model under the current schema). It then reads the file and replaces the read-only model.

import h5py

from pynwb.device import DeviceModel

nwbfile = NWBFile(
    session_description="demonstrate upgrading a legacy device model",
    identifier="NWB123",
    session_start_time=datetime.datetime.now(datetime.timezone.utc),
)
nwbfile.create_device(name="my_device", description="a mass flow controller")

filename = "legacy_device_model.nwb"
with NWBHDF5IO(filename, "w") as io:
    io.write(nwbfile)

# add the legacy Device.model string attribute that schema versions before 2.9 used
with h5py.File(filename, "r+") as f:
    f["general/devices/my_device"].attrs["model"] = "MFC_200/250-0.66_40mm_MF2.5:FLT"

Read the file and replace the read-only model with a new DeviceModel that has a valid name, copying the metadata from the original model. The example builds a valid name by replacing / and : with _.

The legacy file stores model as a string attribute, so it cannot be modified in place (mode="r+") or written to a new path with write. Use export instead, and call set_modified on the Device so that export rewrites it, dropping the legacy string attribute and writing a link to the new DeviceModel.

with NWBHDF5IO(filename, mode="r") as read_io:
    read_nwbfile = read_io.read()
    device = read_nwbfile.devices["my_device"]
    read_only_model = device.model

    # build a valid name by replacing the characters that are not allowed in NWB object names
    valid_name = read_only_model.name.replace("/", "_").replace(":", "_")

    # create a new DeviceModel with a valid name, copying the metadata from the read-only model
    new_model = DeviceModel(
        name=valid_name,
        manufacturer=read_only_model.manufacturer,
        model_number=read_only_model.model_number,
        description=read_only_model.description,
    )

    # Device.model is write-once, so clear it with fields.pop before assigning the new model
    device.fields.pop("model")
    device.model = new_model
    read_nwbfile.add_device_model(new_model)

    # mark the Device as modified so export rewrites it without the legacy string attribute
    device.set_modified(True)

    export_filename = "upgraded_device_model.nwb"
    with NWBHDF5IO(export_filename, mode="w") as export_io:
        export_io.export(src_io=read_io, nwbfile=read_nwbfile)

# the exported file reads back with a valid, writable DeviceModel and no upgrade warning
with NWBHDF5IO(export_filename, "r") as io:
    read_nwbfile = io.read()
    print(read_nwbfile.devices["my_device"].model)

For more information about the export functionality, see Exporting NWB files and the PyNWB documentation for NWBHDF5IO.export.

For more information about editing a file in place, see Editing NWB files.