NeuroTask is a benchmark dataset designed to facilitate the development of accurate and efficient methods for analyzing multi-session, multi-task, and multi-subject neural data. It integrates six datasets from motor cortical regions, comprising 148 sessions from 19 subjects engaged in 7 distinct tasks.
The figure here shows an illustration.
- NeuroTask nwb version was introduced at the Open Neurodata Showcase '24
- Dataset is now also available in NWB format on Dandi and the tutorial.
The public dataset suite is available for download through Dandi in NWB format.
Each dandiset corresponds to a specific task.
Alternatively, you can access the dataset in Parquet format through Kaggle.
-
If you are using the web user interface, you can download all data from the provided link. The download button is at the upper right corner of the webpage.
-
If you would like to use the Kaggle API, please follow the instructions here. After setting the API correctly, you can simply use the command below to download all data.
kaggle datasets download -d carolinafilipe/neurotask-multi-tasks-benchmark-dataset
The dataset schema is comprised of:
- Indexes for dataset,animal, session and trial
- Neurophysiological data.
- Behavior covariates.
- Events time indications.
Each row represents a single time step.
For more information on each of these components, please consult the provided datasheet.
We developed an API, api_neurotask.py, to streamline data loading and preprocessing tasks, such as data rebinning and alignment to specific start event with adjustable offsets. You can find the api tutorial here: tutorial.
In this repository, you will find the code necessary to replicate the experiments of the presented paper.
The first notebook regards the loading, processing and visualization of the data (parquet format) and nwb format .
The second notebook presents the code to reproduce the benchmark results in the manuscript.
NeuroTask has a minimalistic yet powerful API.
To load data in the NeuroTask schema format, you only need a single line of code.
Nwb version:
from api_neurotask import *
data = nap.load_file("001056/sub-Animal-1-&-2/sub-Animal-1-&-2.nwb")
# Retrieve data filtered to include only rewarded trials.
df , bin = get_dataframe(data,filter_result=[b'R'])Parquet version:
from api_neurotask import *
parquet_file_path = 'NeuroTask/2_10_Chowdhury_CObump.parquet'
# Retrieve data filtered to include only rewarded trials.
df,bin = load_and_filter_parquet(parquet_file_path,['A', 'I','F'])The following functions work the same for both versions. You can use the rebin and/or align_event functions as needed.
# Rebin the dataset with a bin size of 20 ms
df = rebin(df,prev_bin_size = bin ,new_bin_size = 20)
# Align each trial of the data (df) to a specific event ('EventTarget_Onset')
# The dataset has a bin size of 20 ms
#We want an offset of -20 ms before the event and 400 ms after the event
df = align_event(df, start_event='EventTarget_Onset', bin_size=20,offset_min=-20,offset_max=400)Prepare the data for model fitting by processing it into input (X) and target (y) variables.
# The input data (X) includes bins of spiking data from before the event, while the target (y) includes bins of future behavior variables (behavior_columns) to predict (e.g., 1 for the next time step prediction).
# The data is split into training, validation, and test sets according to the specified ranges.
# Each position in the resulting lists corresponds to a different session.
X_train_list, y_train_list, X_val_list, y_val_list, X_test_list, y_test_list = process_data(
df,
bins_before=10,
bins_after=1,
training_range=[0, 0.7],
valid_range=[0.7, 0.8],
testing_range=[0.8, 1],
behavior_columns=['hand_vel_x', 'hand_vel_y']
)NeuroTask nwb version works with ...
The NeuroTask benchmark dataset is released under a CC BY-NC 4.0 International License. Our code implementation (api and baselines) is released under the MIT License. We would also like to express our gratitude to the authors of baselines for releasing their code. And to the DANDI, NWB, and CCN software teams for their support and workshops.
Please feel free to contact us if you have any questions about NeuroTask: [email protected]