search

rois-codh / kmnist

Repository for Kuzushiji-MNIST, Kuzushiji-49, and Kuzushiji-Kanji
Creative Commons Attribution Share Alike 4.0 International
693 stars 94 forks source link

Fix kNN benchmark KMNIST accuracy #9

Closed ygabuev closed 4 years ago

ygabuev commented 5 years ago

The benchmark accuracy for kNN classifier with neighbors=4 gives test accuracy of 92.10%, not 91.56% as is currently shown in the results table from README:

C:\...\kmnist\benchmarks> python kuzushiji_mnist_knn.py
Fitting KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
           metric_params=None, n_jobs=-1, n_neighbors=4, p=2,
           weights='distance')
Evaluating KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
           metric_params=None, n_jobs=-1, n_neighbors=4, p=2,
           weights='distance')
Test accuracy: 0.921

Besides, the header of benchmarks/kuzushiji_mnist_knn.py falsely states that the accuracy is 97.4%.

mxbi commented 4 years ago

You're right! This happened due to our minor dataset update in February. Merged.