# Lever Position from Mouse Brain Activity: Data description

*English translation by SOTA – AI Community of the Japanese original, shared under CC BY-NC-SA 3.0 IGO like the original. Organisers who would like this translation removed can email sota.ai.community@gmail.com.*

*Data tab of the Kaggle competition "[Playground]JOAI Competition 2026". On Kaggle the data are 3 CSV files (351.32 MB).*

This dataset was created from experimental data recorded while mice performed an experiment in which they pull a lever to obtain water as a reward; the following were recorded simultaneously:

- experimental information (mouse ID, experiment day, time information)
- brain activity information
- lever state (target)

In this competition, a time series over a short time window (about 20–40 frames) is treated as one set, and the aim is to predict the time series of the "lever position (lever)" from the mouse's behaviour and brain activity.


## Overview of the data

Several mice took part in the experiment, and each mouse underwent 14 days of experiments.
The data are multiple short time series cut out of one day's experiment of a given mouse.
Each cut-out time series can be identified by the ID column sample_id.
One sample_id consists of 20–40 consecutive rows of data (frames).

![dataset_overview](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1305588%2F0f7f485356a2383bf9812c8f88525e54%2Fjoai2026_data_overview.png?generation=1767012459308600&alt=media)

*Figure labels (translated), from top to bottom: "14-day experiment for each mouse"; "one day's experimental record (lever & brain activity)"; "short time-series samples (20–40 frames)"; table columns `lever` (target variable) and `brain` (brain activity); "frames (records) belonging to sample_id: 102"; "linked to mouse_id: 1, day_n: 7".*


## Brain activity (wide-field calcium imaging)

This dataset uses the standardised brain atlas of the Allen Mouse Brain Common Coordinate Framework (Allen CCF), a standard parcellation of the brain provided by a US research institute, to divide the brain into 22 anatomical regions on each of the left and right sides, symmetrically, and stores the mean activity time series of each region. These brain activity data were recorded simultaneously during the mouse behavioural experiment described in the Overview. For details such as the processing methods, please refer to the paper on the dataset (see the Overview).

![allen ccf](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1305588%2Fa0fcfc55f0dde19c69028d05936b0ed7%2F2026-02-03%200.52.03.png?generation=1770047554342231&alt=media)

## Files

`train.csv`

Training data with the lever position (lever).

Each row corresponds to "one frame at a given time" and has the following columns:

- `id`: record-level ID
- `sample_id`: sequence ID  
- `mouse_id`: mouse identifier  
- `day_n`: experiment day number for that mouse  
- `time`: timestamp  
- `lever`: lever position (target variable)  
- `AUDp_l, AUDs_l, ...., VISrl_r`: brain activity information

---

`test.csv`

Test data for submission and score calculation.
Its columns are those of train.csv without the target variable `lever`.  
**The prediction target is this `lever`.**

---

`sample_submission.csv`

A file in the submission format.

- `id`: record-level ID
- `lever`: lever position (target variable)

Because it is a sample, `lever` is 0.0 in every row.