In MNIST challenge, what should be the optimal value of k ?
What should be the ideal value for k?
Hey @ishabehera, the ideal way around to judge optimal value of k is first split the data into 80:20, Now in Knn use first 80% of data, as training data and check accuracy on remaining 20% of data.
K value should be ideally odd (to avoid equal vote contradiction) and K< sqrt(number of training examples).
Now, for different problems you need to run for various values of k and check the score/accuracy on validation data, and than report the best value.
Hope this resolved your doubt.
Plz mark the doubt as resolved in my doubts section.