Kuiz K-Jiran Terdekat

Jawapan dan penjelasan muncul semasa anda mengambil kuiz.

1. Bagaimanakah KNN mengklasifikasikan titik data baharu?
   • Dengan mengambil undi majoriti daripada K jiran terdekatnya
   • Dengan melatih rangkaian saraf di atasnya
   • Dengan mengira garis regresi berwajaran
   • Dengan memberikan kelas rawak
2. Why is KNN often called a “lazy learner”?
   • It never stores training data
   • It stores training examples and can defer much model-specific work until prediction, though indexes may be built during fitting
   • It always trains a deep neural network first
   • It does no work during either fitting or prediction
3. Mengapa penskalaan ciri penting untuk KNN?
   • Ia menjadikan latihan lebih cepat
   • Ia mengurangkan bilangan jiran yang diperlukan
   • Jadi ciri dengan julat yang besar tidak menguasai jarak
   • Ia menukar K secara automatik
4. Apakah yang berlaku jika anda memilih K yang sangat kecil, seperti K=1?
   • Model menjadi sangat licin
   • Model mengabaikan data
   • Ramalan menjadi mustahil
   • Model menjadi sensitif kepada hingar dan mungkin terlalu muat
5. In a brute-force KNN implementation, why can prediction be slow on a large dataset?
   • It may compute distances from a query to many or all stored examples
   • It must update all model weights for every query
   • It reruns the entire training dataset through a neural network
   • It cannot use an index or approximate search
6. Teknologi moden yang manakah bergantung pada idea jiran terdekat?
   • Pemampatan imej dengan JPEG
   • Pencarian vektor dan penjanaan penambahan perolehan
   • Penjanaan nombor rawak
   • Formula hamparan
7. Why should a feature scaler be fit only on the training split before KNN evaluation?
   • To make the distance metric ignore every feature
   • To guarantee all classes have equal counts
   • To avoid using information from validation or test examples when setting the scale
   • To remove the need to choose K
8. How can KNN make a prediction for a regression task?
   • Choose the most common class label only
   • Fit a new deep network for each query
   • Ignore the stored examples and use a fixed intercept
   • Aggregate the target values of the K nearest training examples, often by averaging