Instructions to use Hapyr/her2-lbp-classifier with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Scikit-learn
How to use Hapyr/her2-lbp-classifier with Scikit-learn:
from huggingface_hub import hf_hub_download import joblib model = joblib.load( hf_hub_download("Hapyr/her2-lbp-classifier", "sklearn_model.joblib") ) # only load pickle files from sources you trust # read more about it here https://skops.readthedocs.io/en/stable/persistence.html - Notebooks
- Google Colab
- Kaggle
HER2 LBP Classifier (baseline)
Classificador de imagens histopatologicas HER2 (4 classes: 0, 1+, 2+, 3+)
sobre features LBP (Local Binary Pattern, 10 bins), do projeto
her2-histopathology-classification.
Modelo
StandardScaler + SVC(kernel=rbf, C=10)β selecionado por CV 5-fold (f1_macro = 0.7006) entre 9 candidatos (RF, ExtraTrees, HistGB, LogReg, SVC, MLP) sobre a particao de treino+validacao (593 amostras).- Avaliacao final no conjunto de teste oficial do Warwick (371 amostras):
- accuracy = 0.6119
- f1_macro = 0.6223
Contexto honesto
Este e o baseline LBP do trabalho. O repo original tambem treina modelos sobre features VGG16 (4096 dim.), que alcancaram f1_macro = 0.8096 (VGG16 + RandomForest). As features VGG16 e os pesos das CNNs nao estao disponiveis aqui porque dependem dos imagens do Warwick (licenca academica, redistribuicao restrita).
Uso
import joblib
import numpy as np
from huggingface_hub import hf_hub_download
path = hf_hub_download("Hapyr/her2-lbp-classifier", "her2_lbp_best.joblib")
model = joblib.load(path)
# X: array/DataFrame com 10 features LBP (lbp_0 .. lbp_9), mesma ordem
# da extracao do notebook 01 do projeto original
pred = model.predict(X) # 0, 1, 2, 3 -> '0', '1+', '2+', '3+'
Extracao das features
Veja notebooks/01_feature_extraction_classical_models.ipynb do projeto
her2-histopathology-classification (LBP uniform, P=8, R=1, histograma
normalizado em 10 bins).
Limites
- Dataset Warwick: uso academico; verifique os termos originais antes de reutilizar.
- Desempenho modesto por design (baseline de features classicas); nao e um dispositivo medico.