Why does my KNeighborsClassifier return some empty predictions?

Question

I'm trying to train a classifier to predict digits from photos and I'm using this dataset: https://www.kaggle.com/ardamavi/sign-language-digits-dataset

But when I use the.predict() function it returns for some labels as [0, 0, 0, 0, 0, 0, 0, 0, 0, 0] which does not predict anything. The problem also increases as I increase n_neighbors.

import datasetreader
from sklearn.neighbors import KNeighborsClassifier

X_train, X_test, y_train, y_test = datasetreader.get_dataset(
    '/Sign-Language-Digits-Dataset-master/Dataset')

nsamples, nx, ny = X_train.shape
d2_X_train = X_train.reshape((nsamples,nx*ny))

nsamples, nx, ny = X_test.shape
d2_X_test = X_test.reshape((nsamples,nx*ny))

clf = KNeighborsClassifier(n_neighbors = 5).fit(d2_X_train, y_train)
y_pred = clf.predict(X_test)

# Predicting y
print("Amount of testdata to predict on: ", len(X_test)) # prints 413
print("Actual predicts: ", sum(sum(y_pred))) # prints only 270.0, should be 413

Answer 1

I was able to solve it by rearrange y_train to be a vector with the target names, before I had it as a matrix where the index showed the right answer. Still don't understand why it returned only some empty answers because of that though.