Prediction

Make predictions with the model.

Prediction and probability

The trained pipeline takes new examples directly with raw measurements. predict_proba shows how confident the decision is:

p = model.predict(X_test[:5])
print(p)
print(model.predict_proba(X_test[:2]).round(2))

In KNN the probabilities are the shares of neighbour votes; with 5 neighbours the values are multiples of 0.2.

Finding wrong predictions

With a boolean mask the indices of wrongly predicted examples are found; then by examining those examples' probabilities you learn where the model is undecided:

tahmin = model.predict(X_test)
yanlis = np.where(tahmin != y_test)[0]
print(yanlis)
print(model.predict_proba(X_test[yanlis]).round(2))

Most errors are usually between two neighbouring classes; the confusion matrix confirms this.

Exercise Inspect the errors

Get the predictions for the test set. Print the number of wrong predictions, then the true classes of the wrongly predicted examples (y_test[yanlis]).

Show example solution

import numpy as np
from sklearn.datasets import load_wine
from sklearn.model_selection import train_test_split
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.neighbors import KNeighborsClassifier

X, y = load_wine(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=42, stratify=y)
model = make_pipeline(StandardScaler(), KNeighborsClassifier(n_neighbors=5)).fit(X_train, y_train)

tahmin = model.predict(X_test)
yanlis = np.where(tahmin != y_test)[0]
print(len(yanlis))
print(y_test[yanlis])

Test yourself

With a 5-neighbour KNN, which can't be a class's probability?
- 0.5
- 0.4
- 0.6
- 1.0
Show answer

0.5 — The probability has the form k/5: 0, 0.2, 0.4, 0.6, 0.8, 1.0.

Get the indices of the wrong predictions.

yanlis = np.where(tahmin ____ y_test)[0]

Show answer

yanlis = np.where(tahmin != y_test)[0]

Match the expressions.

model.predict

Class decision
model.predict_proba

Neighbour vote shares
tahmin != y_test

Error mask
np.where(...)[0]

Error indices

Next lesson: Evaluation

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies