Distance normalization scale is off in the HackAI version, possibly causing the excess error in the distance calculation.
The lines
train_distances = (train_distances-np.min(train_distances))/(np.max(train_distances)-np.min(train_distances))
val_distances = (val_distances-np.min(val_distances))/(np.max(val_distances)-np.min(val_distances))
test_distances = (test_distances-np.min(test_distances))/(np.max(test_distances)-np.min(test_distances))
needs to be corrected to
max_distance = np.max(distances)
min_distance = np.min(distances)
train_distances = (train_distances-min_distance)/(max_distance-min_distance)
val_distances = (val_distances-min_distance)/(max_distance-min_distance)
test_distances = (test_distances-min_distance)/(max_distance-min_distance)
in the next version(s) of the model.
Distance normalization scale is off in the HackAI version, possibly causing the excess error in the distance calculation.
The lines
needs to be corrected to
in the next version(s) of the model.