OmarK211/tabular-test

CAD Model Features Dataset (Text-to-CAD Focus)

OmarK211/tabular-test

A tabular dataset of CAD model geometric and functional parameters, containing manually collected samples (original) and synthetic variants (augmented).

Task Overview

The primary predictive target is base_sketch_shape (categorical 2D profile shape) based on 6 input dimensions and properties.

Dataset Structure & Fields

Field Name	Data Type	Role	Description
base_sketch_shape	Categorical	Target	Core 2D profile shape (e.g., circle, rectangle, L-shape)
max_length	Numeric (Float)	Feature	Continuous maximum length dimension
max_width	Numeric (Float)	Feature	Continuous maximum width dimension
max_depth	Numeric (Float)	Feature	Continuous maximum depth dimension
part_function	Categorical	Feature	Functional role of the CAD model
symmetry	Categorical	Feature	Symmetry type (e.g., planar, rotational, none)
material	Categorical	Feature	Intended material composition
source_id, parent_id, second_parent_id	String	Provenance	Unique row ID and original parent source IDs
augmentation, is_augmented, mix_weight	Metadata	Provenance	Augmentation method, synthetic flag, and Mixup interpolation weight

Dataset Splits

Split	Original rows	Synthetic rows	Total rows
train	21	315	336
validation	4	0	4
test	5	0	5

Target Class Distribution (base_sketch_shape)

Split	Base Sketch Shape	Rows
train	amorphous	32
train	circular	160
train	rectangular	144
validation	amorphous	2
validation	circular	1
validation	rectangular	1
test	circular	1
test	rectangular	4

Augmentation Methods

Synthetic samples were generated using 5 copy requests per method:

Augmentation method	Retained synthetic rows
additive_numeric_jitter	105
mixup_interpolation	105
multiplicative_numeric_scale	105

• Additive Numeric Jitter: Applied Gaussian noise to continuous CAD dimensions (max_length, max_width, max_depth), clipped at positive bounds (>0.01).
• Multiplicative Numeric Scaling: Scaled continuous dimensions independently by uniform factors in [0.90, 1.10].
• Mixup Interpolation: Linearly interpolated continuous dimensions between distinct parents using weights in [0.60, 0.90], retaining the primary parent's categorical fields.

Intended use and limitations

Use for teaching/learning mixed-type data contracts, provenance, augmentation, and small-sample regression. This convenience sample does not represent the general population.

Privacy and licensing

No license was specified in repository metadata when this card was first added; this card does not assign one.

Load and compare

from datasets import load_dataset
ds = load_dataset("OmarK211/tabular-test")
# 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.