wowo0709

wowo0709

큰 사람이 되려면 큰 물에서 놀아라

[Machine Learning] Linear Classification 2 (Practice)

3 minute read

Example 1: Binary Classification

Iris dataset을 사용한 binary classification 실습

  • Setosa VS non-Setosa

Import dataset

from sklearn.datasets import load_iris
iris = load_iris()
print(type(iris)) 

<class 'sklearn.utils.Bunch'>

print(iris.feature_names) # 4 features
print(iris.target_names)  # 3 labels

['sepal length (cm)', 'sepal width (cm)', 'petal length (cm)', 'petal width (cm)']
['setosa' 'versicolor' 'virginica']

X, y = iris.data, iris.target
print(X.shape, type(X))
print(y.shape, type(y))

(150, 4) <class 'numpy.ndarray'>
(150,) <class 'numpy.ndarray'>


Make to binary classification dataset

X2 = X[:, :2]   # first two features

%matplotlib inline
import matplotlib.pyplot as plt

markers = ['o', '+', '^']
for i in range(3):
    xs = X2[:, 0][y == i]
    ys = X2[:, 1][y == i]
    plt.scatter(xs, ys, marker=markers[i])
plt.legend(iris.target_names)
plt.xlabel("Sepal length")
plt.ylabel("Sepal width")

Text(0, 0.5, 'Sepal width')

output_8_1

위의 label 분포를 보면 setosa와 versicolor&virginica 로 크게 이진 분류를 할 수 있다.

따라서 label의 virginica를 모두 versicolor로 바꿔서 이진 분류 데이터로 바꾼다.

y2 = y.copy()      # y의 복사본을 만든다
y2[(y2==2)] = 1    # y중에 2의 값을 모두 1로 바꾼다 -> 이진 분류
y2

markers = ['o', '+', '^']
for i in range(3):
    xs = X2[:, 0][y2 == i]
    ys = X2[:, 1][y2 == i]
    plt.scatter(xs, ys, marker=markers[i])
binary_names = ['setosa', 'non-setosa']
plt.legend(binary_names)
plt.xlabel("Sepal length")
plt.ylabel("Sepal width")

Text(0, 0.5, 'Sepal width')

output_10_1


Split train/test dataset

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X2, y2, test_size=0.3, random_state=17)
X_train.shape, X_test.shape, y_train.shape, y_test.shape

((105, 2), (45, 2), (105,), (45,))

markers = ['o', '+', '^']
for i in range(3):
    xs = X_train[:, 0][y_train == i]
    ys = X_train[:, 1][y_train == i]
    plt.scatter(xs, ys, marker=markers[i])
binary_names = ['setosa', 'non-setosa']
plt.legend(binary_names)
plt.xlabel("Sepal length")
plt.ylabel("Sepal width")

Text(0, 0.5, 'Sepal width')

output_13_1


Select Model and Train

from sklearn.linear_model import SGDClassifier
clf = SGDClassifier(max_iter=2000, random_state=42)
clf.fit(X_train, y_train)

SGDClassifier(max_iter=2000, random_state=42)

clf.coef_, clf.intercept_

(array([[  98.14963797, -139.44757308]]), array([-90.4188475]))

a = clf.coef_[0,0]
b = clf.coef_[0,1]
c = clf.intercept_
print(a, b, c)

98.1496379726464 -139.44757307589103 [-90.4188475]


Inference and Plotting

import numpy as np
%matplotlib inline
import matplotlib.pyplot as plt

markers = ['o', '+', '^']
for i in range(2):
    xs = X_train[:, 0][y_train == i]
    ys = X_train[:, 1][y_train == i]
    plt.scatter(xs, ys, marker=markers[i])

binary_names = ['setosa', 'non-setosa']
plt.legend(binary_names)
plt.xlabel("Sepal length")
plt.ylabel("Sepal width")

XX = np.linspace(4, 8, 40)
# 결정 경계선
plt.plot(XX, (-a/b)*XX - c/b, "k-", linewidth=2)

[<matplotlib.lines.Line2D at 0x17c7dd13c70>]

output_19_1

print(clf.predict([[4.5, 3.5]]))  # 0

[0]

print(clf.score(X2, y2))

0.9933333333333333

print(clf.score(X_test, y_test))

1.0


Use KFold and cross_val_score

from sklearn.model_selection import cross_val_score, KFold
from sklearn.linear_model import SGDClassifier

cv = KFold(n_splits=5, shuffle=True)

'''
X2.shape[0]
y

for train_index, test_index in cv.split(X2):
   print("TRAIN:", train_index, "TEST:", test_index)
   X_train, X_test = X2[train_index], X2[test_index]
   y_train, y_test = y2[train_index], y2[test_index]
'''

# 위의 과정을 한 번에 해 주는 함수가 있음.
score = cross_val_score(SGDClassifier(), X2, y2, cv=cv)
print(score, score.mean())

[0.93333333 1.         1.         0.96666667 0.8       ] 0.9400000000000001



Example 2: More difficult classification

Still binary classification.

  • Virginica VS non-Viginica

Make dataset

iris = load_iris()
X, y = iris.data, iris.target
X3 = X[:,[0,2]] # feature selection
y3 = y.copy()
y3[y3 == 1] = 0 # 0, 1 -> 0
y3[y3 == 2] = 1 # 2 -> 1
X_train, X_test, y_train, y_test = train_test_split(X3, y3, test_size=0.3, random_state=17)

markers = ['o', '+', '^']
for i in range(3):
    xs = X_train[:, 0][y_train == i]
    ys = X_train[:, 1][y_train == i]
    plt.scatter(xs, ys, marker=markers[i])
binary_names = ['non-virginica', 'virginica']
plt.legend(binary_names)
plt.xlabel("Sepal length")
plt.ylabel("Petal length")

Text(0, 0.5, 'Petal length')

output_26_1


Model selection and training

첫번째 모델은 alpha 값을 작게 주어 에러를 줄이는 데 더 집중하도록 한다.

clf1 = SGDClassifier(penalty='l2', alpha=0.0001, random_state=42)
clf1.fit(X_train, y_train)
clf1.score(X_test, y_test)

0.9111111111111111


두번째 모델은 alpha 값을 크게 주어 가중치를 줄이는 데 더 집중하도록 한다.

clf2 = SGDClassifier(penalty='l2', alpha=1, random_state=42)
clf2.fit(X_train, y_train)
clf2.score(X_test, y_test)

0.6888888888888889

너무 강하게 가중치를 규제하면 성능이 떨어진다.


Plotting

x0, x1 = np.meshgrid(
        np.linspace(3, 9, 200).reshape(-1, 1),
        np.linspace(1, 8, 200).reshape(-1, 1),
    )
X_new = np.c_[x0.ravel(), x1.ravel()]

y_predict1 = clf1.predict(X_new)
y_predict2 = clf2.predict(X_new)

zz1 = y_predict1.reshape(x0.shape)
zz2 = y_predict2.reshape(x0.shape)

plt.figure(figsize=(12,8))
plt.subplot(2,2,1)

# clf1 -> train
plt.plot(X_train[y_train==1, 0], X_train[y_train==1, 1], "rx", label="virginica")
plt.plot(X_train[y_train==0, 0], X_train[y_train==0, 1], "yo", label="non-virginica")
plt.contourf(x0, x1, zz1)
plt.xlabel("setal length")
plt.ylabel("petal length")
plt.legend(loc="lower right")
# clf1 -> test
plt.subplot(2,2,2)
plt.plot(X_test[y_test==1, 0], X_test[y_test==1, 1], "rx", label="virginica")
plt.plot(X_test[y_test==0, 0], X_test[y_test==0, 1], "yo", label="non-virginica")
plt.contourf(x0, x1, zz1)
plt.xlabel("setal length")
plt.ylabel("petal length")
plt.legend(loc="lower right")
# clf2 -> train
plt.subplot(2,2,3)
plt.plot(X_train[y_train==1, 0], X_train[y_train==1, 1], "rx", label="virginica")
plt.plot(X_train[y_train==0, 0], X_train[y_train==0, 1], "yo", label="non-virginica")
plt.contourf(x0, x1, zz2)
plt.xlabel("setal length")
plt.ylabel("petal length")
plt.legend(loc="lower right")
# clf2 -> test
plt.subplot(2,2,4)
plt.plot(X_test[y_test==1, 0], X_test[y_test==1, 1], "rx", label="virginica")
plt.plot(X_test[y_test==0, 0], X_test[y_test==0, 1], "yo", label="non-virginica")
plt.contourf(x0, x1, zz2)
plt.xlabel("setal length")
plt.ylabel("petal length")
plt.legend(loc="lower right")
plt.show()

output_33_0


clf1.coef_, clf2.coef_

(array([[-151.63660655,  225.78490314]]), array([[-0.07619048,  0.42329004]]))

Leave a comment

You may also enjoy