Reading NWB files

Python libraries

To read NWB files using python you must have the pynwb package.

See here for instructions on installing package.

Opening the File in python

To open the file in python you must first import NWBHDF5IO from the pynwb package.

from pynwb import NWBHDF5IO

---

Then do the following,

with NWBHDF5IO('path/to/nwb/file.nwb', 'r') as io:
    nwbfile = io.read()
    print(nwbfile)

Structure of NWB File

NWB files are basically stored like a list of dictionaries with fields such as acquisition, devices, and electrodes.

---

Acquisition

nwbfile.acquisition

Acquisition holds all the raw datasets.

Each dataset then has additional attributes.

Depending on the task, there are a varying amount of datasets

For example, there are 9 datasets for the sample cst files:

with NWBHDF5IO('/path/to/nwb/file.nwb', 'r') as io:
    nwbfile = io.read()
    print(list(nwbfile.acquisition.keys()))

['ElectricalSeries', 'Eyetrack_Argus', 'Head_Location_Argus', 'Head_Rotation_Argus', 'Monitor_Eyetrack_Argus', 'Pupil_Diameters_Argus', 'StimLabels', 'allCSTdata', 'allOpenSignalsData']

---

ElectricalSeries

nwbfile.acquisition['ElectricialSeries']

The 'ElectricalSeries' dataset holds the information about the eeg data in microvolts.

The relavent attributes this dataset has are:

• .data this holds all the raw data
• .timestamps this holds all the timestamps
• .description this holds the column headers as a string. Split string by ',' to get as a list for use in pandas dataframe
• .electrodes Stores table for detailed information about electrodes such as general location, coordinates, and impedances.

Eyetrack_Argus

nwbfile.acquisition['Eyetrack_Argus'].spatial_series['Eyetrack_Argus']

The 'Eyetrack_Argus' dataset holds information about the data collected from the argus eyetracker, specifically the position of the eye as it moves.

The relavent attributes this dataset has are:

• .data this holds all the raw data
• .timestamps this holds all the timestamps
• .description Description of spatial series

Monitor_Eyetrack_Argus

nwbfile.acquisition['Monitor_Eyetrack_Argus'].spatial_series['Monitor_Eyetrack_Argus']

The 'Monitor_Eyetrack_Argus' dataset holds information about the data collected from the argus eyetracker, specifically where the eye is looking at the monitor.

The relavent attributes this dataset has are:

• .data this holds all the raw data
• .timestamps this holds all the timestamps
• .description Description of spatial series

Head_Location_Argus

nwbfile.acquisition['Head_Location_Argus'].spatial_series['Head_Location_Argus']

The 'Head_Location_Argus' dataset holds information about the data collected from the argus eyetracker, specifically the coordinates in cm of the head as it moves in space. 0,0,0 is located at the tracker. +X is distance from monitor, +Y is Right, +Z is Down'

The relavent attributes this dataset has are:

• .data this holds all the raw data
• .timestamps this holds all the timestamps
• .description Description of spatial series

Head_Rotation_Argus

nwbfile.acquisition['Head_Rotation_Argus']

The 'Head_Rotation_Argus' dataset holds information about the data collected from the argus eyetracker, specifically the rotation of the head as it moves in space. -180,0,0 is the subject looking straight ahead at the monitor.

The relavent attributes this dataset has are:

• .data this holds all the raw data
• .timestamps this holds all the timestamps
• .description this holds the column headers as a string. Split string by ',' to get as a list for use in pandas dataframe

Pupil_Diameters_Argus

nwbfile.acquisition['Pupil_Diameters_Argus']

The 'Pupil_Diameters_Argus' dataset holds information about the data collected from the argus eyetracker, specifically the diameter of the pupil

The relavent attributes this dataset has are:

• .data this holds all the raw data
• .timestamps this holds all the timestamps
• .description this holds the column headers as a string. Split string by ',' to get as a list for use in pandas dataframe

Stimlabels

nwbfile.acquisition['StimLabels']

The 'StimLabels' dataset holds information about StimLabels

The relavent attributes this dataset has are:

• .data this holds all the raw data
• .timestamps this holds all the timestamps

allCSTdata

nwbfile.acquisition['allCSTdata']

The 'allCSTdata' dataset holds information about CST data

The relavent attributes this dataset has are:

• .data this holds all the raw data
• .timestamps this holds all the timestamps
• .description this holds the column headers as a string. Split string by ',' to get as a list for use in pandas dataframe
• .units this holds the units for each column as a string

allOpenSignalsData

nwbfile.acquisition['allOpenSignalsData']

The 'allOpenSignalsData' dataset holds information about general biological data

The relavent attributes this dataset has are:

• .data this holds all the raw data
• .timestamps this holds all the timestamps
• .description this holds the column headers as a string. Split string by ',' to get as a list for use in pandas dataframe
• .units this holds the units for each column as a string

Mindlogger

nwbfile.acquisition['Mindlogger']

The 'Mindlogger' dataset holds information about MindLogger data

The relavent attributes this dataset has are:

• .data this holds all the raw data
• .timestamps this holds all the timestamps
• .description this holds the column headers as a string. Split string by ',' to get as a list for use in pandas dataframe

---

Loading Data

You might see this output when loading .data,

<HDF5 dataset "data": shape (397540, 64), type "<f4">

This is the object that is loaded by pynwb to be indexed. To load the entire array into memory, just use [:] at the end.

All raw data are stored as numpy arrays.

---

Creating pandas dataframe from data

Since some data is loaded together with column headers in the description, you can create a pandas dataframe with everything provided. For example, here is a snippet of code to turn 'allCSTdata' into a pandas dataframe:

from pynwb import NWBHDF5IO
import pandas as pd

with NWBHDF5IO('/path/to/nwb/file.nwb', 'r') as io:
    nwbfile = io.read()
    # Data Array
    cst = nwbfile.acquisition['allCSTdata'].data[:]
    # Timestamps for each data point
    times = nwbfile.acquisition['allCSTdata'].timestamps[:]
    # Description for cst data holds all the headers
    headers = nwbfile.acquisition['allCSTdata'].description.split(',')
    # The units for each column, not used in dataframe but can be called for later use
    units = nwbfile.acquisition['allCSTdata'].unit.split(',')

# Creating Dataframe
df = pd.DataFrame(cst, columns=headers)
# Adding timestamps as a column
df['times'] = times
print(df)

pynwb Documentation

For more information the pynwb documentation can be found here

NWB Documentation

For more information about the NWB file format, the documentation can be found here