BloublouWhy/sign-mnist-mlp

Sign Language MNIST — MLP Classifier

• Dataset: datamunge/sign-language-mnist via kagglehub (27455 train / 7172 test, 28x28 grayscale, 24 classes, labels 0-24 excluding 9/J and 25/Z).
• Preprocessing: pixels / 255.0, no reshape (784-dim vectors).
• Model: sklearn.neural_network.MLPClassifier(hidden_layer_sizes=(256,128), activation='relu', max_iter=30, random_state=42).
• Test accuracy: 0.7886.
• Artifact: sign_mnist_mlp.pkl (joblib).

Usage

import joblib
model = joblib.load("sign_mnist_mlp.pkl")
proba = model.predict(X_test_scaled)  # X scaled /255.0, shape (n, 784)

Publish status

Published locally as sign_mnist_mlp.pkl in this repo (no Hugging Face token configured, so no remote push). To push to the Hub: huggingface-cli login then huggingface-cli upload <repo> sign_mnist_mlp.pkl.