머신러닝(Machine Learning) - 지도학습(3)

import pandas as pd
import numpy as np

fish = pd.read_csv('https://bit.ly/fish_csv_data')
fish.head()

print(pd.unique(fish['Species']))

['Bream' 'Roach' 'Whitefish' 'Parkki' 'Perch' 'Pike' 'Smelt']

fish_input = fish[['Weight', 'Length', 'Diagonal', 'Height', 'Width']].to_numpy()
# fish_input = fish[fish.columns[1:]].to_numpy()

print(fish_input[:5])

[[242.      25.4     30.      11.52     4.02  ]
 [290.      26.3     31.2     12.48     4.3056]
 [340.      26.5     31.1     12.3778   4.6961]
 [363.      29.      33.5     12.73     4.4555]
 [430.      29.      34.      12.444    5.134 ]]

fish_target = fish['Species'].to_numpy()

fish_target

array(['Bream', 'Bream', 'Bream', 'Bream', 'Bream', 'Bream', 'Bream',
       'Bream', 'Bream', 'Bream', 'Bream', 'Bream', 'Bream', 'Bream',
       'Bream', 'Bream', 'Bream', 'Bream', 'Bream', 'Bream', 'Bream',
       'Bream', 'Bream', 'Bream', 'Bream', 'Bream', 'Bream', 'Bream',
       'Bream', 'Bream', 'Bream', 'Bream', 'Bream', 'Bream', 'Bream',
       'Roach', 'Roach', 'Roach', 'Roach', 'Roach', 'Roach', 'Roach',
       'Roach', 'Roach', 'Roach', 'Roach', 'Roach', 'Roach', 'Roach',
       'Roach', 'Roach', 'Roach', 'Roach', 'Roach', 'Roach', 'Whitefish',
       'Whitefish', 'Whitefish', 'Whitefish', 'Whitefish', 'Whitefish',
       'Parkki', 'Parkki', 'Parkki', 'Parkki', 'Parkki', 'Parkki',
       'Parkki', 'Parkki', 'Parkki', 'Parkki', 'Parkki', 'Perch', 'Perch',
       'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch',
       'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch',
       'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch',
       'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch',
       'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch',
       'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch',
       'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Perch',
       'Perch', 'Perch', 'Perch', 'Perch', 'Perch', 'Pike', 'Pike',
       'Pike', 'Pike', 'Pike', 'Pike', 'Pike', 'Pike', 'Pike', 'Pike',
       'Pike', 'Pike', 'Pike', 'Pike', 'Pike', 'Pike', 'Pike', 'Smelt',
       'Smelt', 'Smelt', 'Smelt', 'Smelt', 'Smelt', 'Smelt', 'Smelt',
       'Smelt', 'Smelt', 'Smelt', 'Smelt', 'Smelt', 'Smelt'], dtype=object)

from sklearn.model_selection import train_test_split

train_input, test_input, train_target, test_target = train_test_split(fish_input, fish_target, random_state=42)

from sklearn.preprocessing import StandardScaler

ss = StandardScaler()
ss.fit(fish_input)
train_scaled = ss.transform(train_input)
test_scaled = ss.transform(test_input)

from sklearn.neighbors import KNeighborsClassifier

kn = KNeighborsClassifier(n_neighbors=3)
kn.fit(train_scaled, train_target)

print(kn.score(train_scaled, train_target))
print(kn.score(test_scaled, test_target))

0.8907563025210085
0.85

# 아스키 코드 순서로 출력

print(kn.classes_)

['Bream' 'Parkki' 'Perch' 'Pike' 'Roach' 'Smelt' 'Whitefish']

print(kn.predict(test_scaled[:5]))

['Perch' 'Smelt' 'Pike' 'Perch' 'Perch']

# predict_proba : 각 클래스별 확률값을 반환해 주는 메서드
# 각 행은 샘플 순서, 각 열은 classes_ 메서드로 반환되는 클래스 순서를 따름

proba = kn.predict_proba(test_scaled[:5])
print(np.round(proba, decimals=4))

[[0.     0.     1.     0.     0.     0.     0.    ]
 [0.     0.     0.     0.     0.     1.     0.    ]
 [0.     0.     0.     1.     0.     0.     0.    ]
 [0.     0.     0.6667 0.     0.3333 0.     0.    ]
 [0.     0.     0.6667 0.     0.3333 0.     0.    ]]

distances, indexes = kn.kneighbors(test_scaled[3:4])
print(train_target[indexes])

[['Roach' 'Perch' 'Perch']]

• 로지스틱 회귀
  : 분류 알고리즘의 기본. 딥러닝에서 배울 인공신경망의 기초가 되는 알고리즘

import matplotlib.pyplot as plt

z = np.arange(-5,5,0.1)
phi = 1 / (1 + np.exp(-z))  # 시그모이드 함수

plt.plot(z, phi)
plt.xlabel('z')
plt.ylabel('phi')
plt.show()

name = np.array(['foo','bar','baz','quz','quux'])
print(name)

['foo' 'bar' 'baz' 'quz' 'quux']

bol = np.array([True, False, True, True, False])
print(bol)

[ True False  True  True False]

name[bol]

array(['foo', 'baz', 'quz'], dtype='<U4')

char_arr = np.array(['A','B','C','D','E'])
print(char_arr[[True, False, True, False, False]])

['A' 'C']

# 분리형 인덱싱

bream_smelt_indexes = (train_target == 'Bream') | (train_target == 'Smelt')
train_bream_smelt = train_scaled[bream_smelt_indexes]
target_bream_smelt = train_target[bream_smelt_indexes]

from sklearn.linear_model import LogisticRegression

lr = LogisticRegression()
lr.fit(train_bream_smelt, target_bream_smelt)

LogisticRegression()

LogisticRegression()

print(lr.predict(train_bream_smelt[:5]))
print(lr.predict_proba(train_bream_smelt[:5]))
# [Bream일 확률   Smelt일 확률]

['Bream' 'Smelt' 'Bream' 'Bream' 'Bream']
[[0.99746125 0.00253875]
 [0.02849095 0.97150905]
 [0.99461153 0.00538847]
 [0.98515891 0.01484109]
 [0.99754417 0.00245583]]

print(lr.classes_)

['Bream' 'Smelt']

print(lr.coef_, lr.intercept_)  # 각 특성에 곱해지는 가중치, 절편

# z = a*무게 + b*길이 + c*대각선 + d*높이 + e*두께 + f

[[-0.42060794 -0.60027616 -0.68776093 -1.00279303 -0.74505586]] [-2.19130847]

# decision_function: 양성 클래스에 대한 z값을 계산

decisions = lr.decision_function(train_bream_smelt[:5])
print(decisions)

[-5.97354273  3.52926408 -5.21809007 -4.19540298 -6.00682974]

# 시그모이드

from scipy.special import expit
print(expit(decisions))

[0.00253875 0.97150905 0.00538847 0.01484109 0.00245583]

print(lr.predict(train_bream_smelt[:5]))

['Bream' 'Smelt' 'Bream' 'Bream' 'Bream']

# 로지스틱 회귀로 다중 분류 수행하기

lr = LogisticRegression(C=20, max_iter=1000)  # C: 규제값 - 클수록 완화
lr.fit(train_scaled, train_target)

print(lr.score(train_scaled, train_target))
print(lr.score(test_scaled, test_target))

0.9243697478991597
0.925

print(lr.predict(test_scaled[:5]))

['Perch' 'Smelt' 'Pike' 'Roach' 'Perch']

proba = lr.predict_proba(test_scaled[:5])
print(np.round(proba, decimals=3))

[[0.    0.015 0.84  0.    0.136 0.007 0.003]
 [0.    0.003 0.046 0.    0.008 0.944 0.   ]
 [0.    0.    0.035 0.935 0.017 0.014 0.   ]
 [0.012 0.034 0.316 0.007 0.556 0.    0.075]
 [0.    0.    0.899 0.002 0.094 0.002 0.001]]

print(lr.classes_)

['Bream' 'Parkki' 'Perch' 'Pike' 'Roach' 'Smelt' 'Whitefish']

print(lr.coef_.shape, lr.intercept_.shape)

(7, 5) (7,)

# 각각의 z 값
decision = lr.decision_function(test_scaled[:5])
print(np.round(decision, decimals=2))

[[ -6.26   1.05   5.1   -2.83   3.27   0.33  -0.66]
 [-10.49   1.95   4.69  -2.58   2.92   7.72  -4.22]
 [ -4.03  -6.13   3.08   6.37   2.34   2.16  -3.78]
 [ -0.62   0.41   2.63  -1.2    3.19  -5.6    1.19]
 [ -6.11  -1.92   5.71  -0.24   3.45  -0.2   -0.69]]

# 소프트 맥수 함수

from scipy.special import softmax
proba = softmax(decision, axis=1)
print(np.round(proba, decimals=3))

[[0.    0.015 0.84  0.    0.136 0.007 0.003]
 [0.    0.003 0.046 0.    0.008 0.944 0.   ]
 [0.    0.    0.035 0.935 0.017 0.014 0.   ]
 [0.012 0.034 0.316 0.007 0.556 0.    0.075]
 [0.    0.    0.899 0.002 0.094 0.002 0.001]]

# 각 클래스에 속할 확률

proba = lr.predict_proba(test_scaled[:5])
print(np.round(proba, decimals=3))

[[0.    0.015 0.84  0.    0.136 0.007 0.003]
 [0.    0.003 0.046 0.    0.008 0.944 0.   ]
 [0.    0.    0.035 0.935 0.017 0.014 0.   ]
 [0.012 0.034 0.316 0.007 0.556 0.    0.075]
 [0.    0.    0.899 0.002 0.094 0.002 0.001]]

# 확률적 경사 하강법

로지스틱 손실 함수 = 이진 크로스엔트로피 손실 함수

타겟이 양성일때는 그대로, 음성일때는 1에서 뺌 ---> 마이너스

양성클래스가 될 확률 - 로지스틱

import pandas as pd

fish = pd.read_csv('https://bit.ly/fish_csv_data')
fish.head()

fish_input = fish[['Weight','Length','Diagonal','Height','Width']].to_numpy()
fish_target = fish['Species'].to_numpy()

from sklearn.model_selection import train_test_split

train_input, test_input, train_target, test_target = train_test_split(fish_input, fish_target, random_state=42)

from sklearn.preprocessing import StandardScaler

ss = StandardScaler()
ss.fit(train_input)
train_scaled = ss.transform(train_input)
test_scaled = ss.transform(test_input)

# 확률적 경사 하강법 - 점진적 학습이 필요할 때
from sklearn.linear_model import SGDClassifier

# loss 손실함수 - log_loss 로그 적용, max_iter 수행할 에포크 횟수(에포크: 한 사이클)
sc = SGDClassifier(loss='log_loss', max_iter=10, random_state=42)
sc.fit(train_scaled, train_target)

print(sc.score(train_scaled, train_target))
print(sc.score(test_scaled, test_target))
# 테스트가 더 높음 - 과소적합
# 경고: 에포크 늘려라

0.773109243697479
0.775


c:\ProgramData\anaconda3\envs\ML\Lib\site-packages\sklearn\linear_model\_stochastic_gradient.py:744: ConvergenceWarning: Maximum number of iteration reached before convergence. Consider increasing max_iter to improve the fit.
  warnings.warn(

# partial_fit: 에포크를 한 번씩 추가로 이어서 훈련 가능
sc.partial_fit(train_scaled, train_target)

print(sc.score(train_scaled, train_target))
print(sc.score(test_scaled, test_target))

0.8151260504201681
0.85

import numpy as np

sc = SGDClassifier(loss='log_loss', random_state=42)

train_score = []
test_score = []

classes = np.unique(train_target)

for _ in range(0,300):
    # 클래스 종류에 뭐가 있는지 미리 알려줘야 실행 가능
    sc.partial_fit(train_scaled, train_target, classes=classes)

    train_score.append(sc.score(train_scaled, train_target))
    test_score.append(sc.score(test_scaled, test_target))

import matplotlib.pyplot as plt

plt.plot(train_score)
plt.plot(test_score)
plt.xlabel('epoch')
plt.ylabel('accuracy')
plt.show()

# tol: 이전 에포크와 이번 에포크의 차이가 tol이 지정해준 실수보다 작으면 반복을 멈춰라
sc = SGDClassifier(loss='log_loss', max_iter=100, tol=None, random_state=42)
sc.fit(train_scaled, train_target)

print(sc.score(train_scaled, train_target))
print(sc.score(test_scaled, test_target))

0.957983193277311
0.925

sc = SGDClassifier(loss='hinge', max_iter=100, tol=None, random_state=42)
sc.fit(train_scaled, train_target)

print(sc.score(train_scaled, train_target))
print(sc.score(test_scaled, test_target))

0.9495798319327731
0.925

선형 - 로지스틱/ 비선형 - svm