Challenge 2

Process data with NumPy and Pandas.

Challenge 2: write KNN from scratch

The whole logic of KNN is a few lines: compute the distances, pick the k nearest examples, take the majority of their labels.

def knn_tahmin(X_egitim, y_egitim, q, k):
    d = np.sqrt(((X_egitim - q) ** 2).sum(axis=1))
    yakin = np.argsort(d)[:k]
    return np.bincount(y_egitim[yakin]).argmax()

Comparing your implementation with scikit-learn's shows that you really understand the algorithm.

Exercise Your own KNN

Write the knn_tahmin(X_egitim, y_egitim, q, k) function. Produce a prediction with k=5 for every example of the Iris test set and print the accuracy with round(..., 3).

Show example solution

import numpy as np
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split

X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=0, stratify=y)

def knn_tahmin(X_egitim, y_egitim, q, k):
    d = np.sqrt(((X_egitim - q) ** 2).sum(axis=1))
    yakin = np.argsort(d)[:k]
    return np.bincount(y_egitim[yakin]).argmax()

p = np.array([knn_tahmin(X_train, y_train, q, 5) for q in X_test])
print(round((p == y_test).mean(), 3))

Exercise Compare with scikit-learn

Train a KNeighborsClassifier(n_neighbors=5) on the same split. Print on one line in how many examples your predictions (p) match its predictions, using (p == m.predict(X_test)).sum(), and the test set size.

Show example solution

import numpy as np
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.neighbors import KNeighborsClassifier

X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=0, stratify=y)
d = np.sqrt(((X_train[None, :, :] - X_test[:, None, :]) ** 2).sum(axis=2))
p = np.array([np.bincount(y_train[np.argsort(r)[:5]]).argmax() for r in d])

m = KNeighborsClassifier(n_neighbors=5).fit(X_train, y_train)
print((p == m.predict(X_test)).sum(), len(X_test))

Test yourself

What does <code>np.argsort(d)[:k]</code> give?
- The indices of the k smallest distances
- The k largest distances
- The k-th distance
- The sorted distances
Show answer

The indices of the k smallest distances — argsort gives indices from smallest to largest; the first k are the nearest neighbours.

Take the majority of the neighbour labels.

sinif = np.bincount(y_egitim[yakin]).____()

Show answer

sinif = np.bincount(y_egitim[yakin]).argmax()

Match the steps.

((X - q) ** 2).sum(axis=1)

Squared distances
np.argsort(d)[:k]

Nearest k
np.bincount

Vote count
argmax

Winning class

Next lesson: Challenge 3

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies