cmuchancel/hw1-tabular-hand-data

24-679 (Fall 2026): CMU Right-Hand Measurements

cmuchancel/hw1-tabular-hand-data

Right-hand finger measurements from Carnegie Mellon University students, plus explicitly marked synthetic training variants. The classroom regression task predicts middle-finger length from thumb, index-finger, ring-finger, and pinky lengths together with the recorded Female / Male categorical feature. All finger measurements are stored in centimeters.

Source and task

The original dataset contains right-hand measurements from 30 CMU students. Each participant measured their own finger lengths in my presence using the same ruler. Lengths were measured from fingertip to the web between fingers and recorded in centimeters. Using the same ruler and measurement instructions helped maintain consistency, although differences in ruler placement and endpoint selection may introduce measurement error.

The preparation notebook reads the hand-measurement CSV, assigns source IDs in retained row order, and checks that finger lengths are finite and positive. The regression target is Middle Finger Size. The four other finger measurements are continuous numeric predictors, and Female / Male is retained as one categorical predictor.

Course: 24-679, Fall 2026, Carnegie Mellon University. Repository maintainer: the account shown above.

Fields

Stored field / group	Meaning and modeling role
Middle Finger Size	Continuous regression target measured in centimeters. Mixup can introduce fractional values.
Thumb Size	Continuous predictor measured in centimeters.
Index Finger Size	Continuous predictor measured in centimeters.
Ring Finger Size	Continuous predictor measured in centimeters.
Pinky Size	Continuous predictor measured in centimeters.
Female / Male	Categorical predictor. Synthetic rows inherit this value from their primary parent.
source_id, parent_id, second_parent_id	Unique example key and original source keys; provenance only.
augmentation, is_augmented, mix_weight	Method, synthetic flag, and primary-parent weight; exclude from predictors.

The machine-readable feature metadata at the top of this card preserves every exact column name and storage type.

Splits and original-source counts

These counts are computed from the packaged splits for this run.

Split	Original rows	Synthetic rows	Total rows
train	21	504	525
validation	4	0	4
test	5	0	5

Requested holdout fraction: 30%; test receives 50% of that holdout. Small-sample rounding changes the realized proportions. The first split uses seed 24679, and the holdout split uses seed 24680.

Original rows are randomly split before augmentation, then both holdout partitions are retained unchanged. Both parents of every Mixup row must come from the original training partition. IDs are assigned from row order; changing or reordering the source CSV can change regenerated IDs and splits. Keep these prepared boundaries fixed for downstream model comparisons.

Augmentation and preprocessing

This run requests 8 copies per method and original training row before filtering.

Every draw starts from original training rows. The copy-count setting requests the same number of draws per method and source; unchanged rows and repeated feature/target combinations for a given primary parent are removed.

• Additive numeric jitter: perturb the four numeric predictor measurements with Gaussian noise. Each standard deviation is max(5% of that training column's IQR, 0.02 cm); clip below zero and round to two decimal places.
• Multiplicative numeric scaling: independently multiply the four numeric predictors by factors in 0.98–1.02, clip below zero, and round to two decimal places.
• Regression Mixup: select two distinct original training parents. Apply the same primary weight in 0.60–0.90 to the four numeric predictors and the middle-finger target. Continuous values are rounded to two decimal places. Copy the primary parent's categorical value and record both parents and the weight.

Jitter and scaling keep the target unchanged. Original and single-parent rows repeat the primary key in second_parent_id and use mix_weight=1.0.

Training method	Stored rows
additive_numeric_jitter	168
multiplicative_numeric_scale	168
none	21
regression_mixup	168

Intended use and limitations

Use for teaching tabular data contracts, provenance, augmentation, and small-sample regression. This convenience sample of 30 CMU students does not represent the general population. Measurements may contain human measurement error, and the small sample size limits conclusions about relationships between hand dimensions.

Synthetic rows are derived from existing training measurements and do not represent additional people. Mixup assumes that interpolation between two measured hands produces a useful synthetic example. More rows do not add independent participants or establish better generalization. Compare original-only and augmented training using the same validation and test sets. Do not use this classroom sample for individual identification or consequential decisions.

Ethical notes

The dataset contains physical measurements and a recorded Female / Male category. These fields should not be interpreted as establishing biological differences or used to make demographic claims from this small convenience sample. The dataset is intended for coursework and demonstration of data preparation and augmentation methods.

Privacy and licensing

The released dataset does not include names, Andrew IDs, or other direct student identifiers. Generated source IDs are used only for provenance and do not correspond to student identifiers. The raw CSV remains a separate source artifact.

Review applicable course permissions before reuse or redistribution. No license is assigned by this card.

AI usage disclosure

ChatGPT was used to help adapt the provided 24-679 lecture notebook to this hand-measurement dataset. The 30 original hand measurements were collected from CMU students and were not generated by AI. Synthetic samples are produced by the explicit augmentation procedures shown in the preparation notebook.

Load and compare

from datasets import load_dataset
ds = load_dataset("cmuchancel/hw1-tabular-hand-data")
# Train with ds["train"], choose settings with ds["validation"], then score ds["test"].

Use an account with access if repository visibility changes. For reproducible comparisons, record the dataset commit and model/environment versions. Regenerate this card with the preparation notebook after changing the data; its counts are calculated from the actual packaged splits. The YAML schema and split configuration are preserved from the upload.