greydanus/mnist1d
A 1D analogue of the MNIST dataset for measuring spatial biases and answering Science of Deep Learning questions.
Constructs synthetic digit templates via procedural generation with configurable transformations (translation, shear, corr noise, resolution), enabling controlled ablation studies of model inductive biases—models with spatial priors (CNNs) achieve 94% accuracy while MLPs plateau at 68%, versus near-parity on standard MNIST. The dataset integrates with NumPy, PyTorch, and TensorFlow ecosystems, available via pip installation or HuggingFace, with Jupyter notebooks demonstrating applications from lottery ticket discovery to metalearning and double descent phenomena.
238 stars and 1,424 monthly downloads. Used by 1 other package. No commits in the last 6 months. Available on PyPI.
Stars
238
Forks
38
Language
Jupyter Notebook
License
Apache-2.0
Category
Last pushed
Oct 09, 2024
Monthly downloads
1,424
Commits (30d)
0
Dependencies
4
Reverse dependents
1
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/greydanus/mnist1d"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Related frameworks
potterhsu/SVHNClassifier
A TensorFlow implementation of Multi-digit Number Recognition from Street View Imagery using...
potterhsu/SVHNClassifier-PyTorch
A PyTorch implementation of Multi-digit Number Recognition from Street View Imagery using Deep...
kensanata/numbers
Handwritten digits, a bit like the MNIST dataset.
kdexd/digit-classifier
A single handwritten digit classifier, using the MNIST dataset. Pure Numpy.
gsurma/digit_recognizer
CNN digit recognizer implemented in Keras Notebook, Kaggle/MNIST (0.995).