Skip to content

Data exploration with python

Jump to bottom
Guy Junior Calvet edited this page Feb 11, 2026 · 1 revision

Keycube Project: Exploratory Data Analysis Report

1. Project Overview

The Keycube is a tangible, cubic-shaped text entry device designed for virtual and mixed-reality (VR/AR) workspaces. By shifting from a traditional 2D plane to a 3D cube, the device offers high mobility and freedom of posture, allowing users to type while standing, sitting, or moving without the need for a desk.

This report analyzes the experimental data from User Study 2 (p=22) to understand the ergonomic relationship between human hand morphology and the cube's 3D geometry.

2. Data Preparation

To analyze the data, we first had to transform the wide-format CSV files into a long-format structure suitable for statistical grouping.

import pandas as pd
import numpy as np
import re

# Load datasets
reachability = pd.read_csv('reachability.csv')
preferences = pd.read_csv('preferences.csv')

# Mapping numeric finger codes to anatomical names
finger_map = {
    1: 'LL', 2: 'LR', 3: 'LM', 4: 'LI', 5: 'LT',
    6: 'RT', 7: 'RI', 8: 'RM', 9: 'RR', 10: 'RL'
}

# Preprocessing Preferences: Transforming keys into a searchable format
pref_keys = [c for c in preferences.columns if len(c) <= 3 and c not in ['Number', 'Handedness']]
pref_long = preferences.melt(id_vars=['Number'], value_vars=pref_keys, 
                             var_name='Key', value_name='FingerNum')
pref_long['FingerCode'] = pref_long['FingerNum'].map(finger_map)
pref_long['Face'] = pref_long['Key'].str[0]

3. Analysis: Finger Preference Heatmap

We visualized which fingers users naturally assign to each face of the cube. This helps identify the "Natural Home Row" for a cubic interface.

# Grouping by Face and Finger to find dominant usage patterns
face_pref = pref_long.groupby(['Face', 'FingerCode']).size().unstack(fill_value=0)

# Visualized via Seaborn Heatmap
# (See viz_exports/Dominant Finger Usage per Cube Face.png)

Key Insight: Users demonstrate a highly consistent "cluster" behavior. The Thumbs (LT/RT) are prioritized for faces that provide structural support, while the Index and Middle fingers navigate the high-frequency side keys.

4. Spatial Reachability (4x4 Grid Layout)

The core of the Keycube is its 4x4 grid on each face. To analyze reachability accurately, we developed a function to extract key numbers using Regular Expressions to avoid conflicts with finger names (e.g., distinguishing the "R" in "Red Face" from the "R" in "Right Ring finger").

def get_grid_data(reach_df, face_code):
    # Use regex to find keys like -R1, -R16 while ignoring finger names
    regex_pattern = rf'-{face_code}\d+$'
    cols = [c for c in reach_df.columns if re.search(regex_pattern, c)]
    
    data = reach_df[cols].mean().reset_index()
    data.columns = ['FingerKey', 'Score']
    
    # Extract only the key number from the end of the string
    data['KeyNum'] = data['FingerKey'].apply(lambda x: int(re.findall(r'\d+', x)[-1]))
    
    # Pivot to a 4x4 physical representation
    return data.groupby('KeyNum')['Score'].mean().sort_index().values.reshape(4, 4)

Ergonomic Result: The heatmaps generated by this code (stored as Average Reachability per Key.png) show that outer edges and corners suffer from a significantly lower reachability score compared to the center-top keys.

5. Hand Ergonomics & Span Correlation

A critical part of the exploration was determining if the Keycube design is "Hand-Size Agnostic." We correlated the users' Hand Span with their Average Reachability Score.

# Correlating physical span with ease of use
reach_cols = [c for c in reachability.columns if '-' in c]
reachability['AvgReach'] = reachability[reach_cols].mean(axis=1)

# Linear regression plot to check correlation
# (See viz_exports/Does Hand Span Predict Reachability.png)

Finding: The data reveals a positive correlation between hand span and total reachability. This suggests that users with a smaller span (< 190mm) struggle with the current cube dimensions, highlighting a need for a "Mini" version of the hardware.

6. Conclusion and Future Directions

The 2D EDA performed here validates the primary ergonomic theories of the Keycube project:

  1. Preference follows Ergonomics: Users almost exclusively choose fingers that have a reachability score $&gt; 1.0$ for a given key.
  2. Physical Constraints: Corner keys are statistically harder to reach across all demographics.

Why 3D is the Next Step

While these Python visualizations confirm the statistical trends, they cannot show the occlusion or the joint angles required to reach a specific key.

3D Visualization is the final frontier for this data. By mapping these 2D reachability averages onto a 3D OBJ model of the Keycube, we can provide developers with a real-time "Ergonomic Heat-Map" that moves as the virtual hand moves.

Clone this wiki locally