Kaggle
DataAnalysis sklearn_TypeB
PCA- Feature Dimenssion Reduction 다뤄보자
2021-06-28 10:10
sklearn_TypeB
- AUTHOR: SungwookLE
- DATE: ‘21.6/28
문제:
- 5차원으로 데이터의 차원 축소
- 축소를 통해 정리된 인자들의 특성을 살펴라
OVEVIEW
- Data Load and View
- Feature Dimenstion Reduce
- Feature Character
1. Data Load and View
from subprocess import check_output
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
import seaborn as sns
sns.set()
from sklearn.preprocessing import StandardScaler
print(check_output(["ls","input"]).decode('utf8'))
AI경진대회 예선(B형)_data-set.zip
input.csv
output.csv
input = pd.read_csv('input/input.csv', header=None)
output = pd.read_csv('input/output.csv', header=None)
output = output.rename(columns={0:'Label'})
input.values
array([[ 0.83035958, -0.33025241, -0.23054277, ..., -1.02979077,
-4.27514811, -0.59929727],
[ -0.04399859, 0.22065793, 1.60051901, ..., -1.10753423,
-20.25542908, -0.56636377],
[ 0.62671752, 2.10042501, -0.96579802, ..., -1.03976259,
-10.22693074, -1.05338458],
...,
[ -0.71817134, 0.26945901, 0.53723753, ..., 0.44234589,
-13.406614 , 0.85427125],
[ -0.3884856 , -0.20375512, 1.40039956, ..., -1.08230872,
40.66522873, -1.58154278],
[ -0.09540666, 1.47321441, 1.05998807, ..., -1.11836725,
27.74371615, -1.51622948]])
# Standard Scaling: Data Normalization -> set mean 0 , and std 1.
stscaler= StandardScaler()
stscaler.fit(input)
input_ = stscaler.transform(input)
y = output
data = pd.DataFrame(input_)
data.head()
| 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 0.818106 | -0.325165 | -0.246214 | -0.012640 | -0.736274 | 1.205771 | -1.074941 | 0.554851 | 1.166372 | 0.445609 | 1.065898 | 0.114682 | -0.506004 | 0.006854 | 1.423487 | -0.674604 | 0.299847 | -1.193585 | -0.215506 | -0.523844 |
| 1 | -0.051535 | 0.224741 | 1.586533 | -0.806663 | -1.905261 | -1.428647 | -1.138761 | -0.189653 | 1.009079 | 1.932479 | 0.060460 | 1.766632 | 1.395304 | -0.959874 | 0.196900 | -0.570431 | 0.004655 | -1.283734 | -1.005080 | -0.494709 |
| 2 | 0.615563 | 2.101084 | -0.982146 | -0.264738 | 1.043126 | 2.171876 | -1.185627 | 0.103988 | 0.612144 | -0.328839 | 0.849970 | -0.459181 | -1.007995 | 1.222779 | -0.832886 | 0.692303 | -1.271343 | -1.205148 | -0.509579 | -0.925548 |
| 3 | 0.607127 | -2.041983 | 0.181954 | -0.300708 | -1.254648 | -2.385617 | -0.814498 | -0.312310 | 1.381110 | -0.885105 | -0.578846 | -0.599110 | 0.880172 | -1.801167 | 0.113969 | 0.472810 | -0.432572 | -0.787749 | 1.030029 | -0.761267 |
| 4 | 1.416870 | 0.361684 | 0.424445 | -1.209425 | 0.349353 | -0.397125 | -0.460240 | 0.267333 | -1.291405 | 1.225944 | 1.432778 | 0.157923 | -0.985794 | 0.625482 | -0.446556 | -2.052565 | -0.418550 | -0.468195 | 1.015300 | -0.358104 |
data.describe()
| 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| count | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 | 1.000000e+04 |
| mean | -3.087669e-17 | -1.141309e-17 | 9.348078e-18 | -5.060535e-17 | 3.215206e-17 | 3.910205e-17 | -1.594280e-17 | -4.818368e-18 | 6.561418e-18 | 2.553513e-17 | 7.327472e-19 | -2.753353e-18 | -3.052072e-18 | -3.863576e-18 | -3.530856e-17 | -3.313044e-17 | 3.850253e-17 | 1.809664e-17 | -4.207745e-17 | 4.478640e-17 |
| std | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 | 1.000050e+00 |
| min | -3.688757e+00 | -3.859865e+00 | -4.714368e+00 | -3.885379e+00 | -3.629213e+00 | -3.770522e+00 | -1.870203e+00 | -3.621609e+00 | -3.600433e+00 | -3.997427e+00 | -3.868648e+00 | -3.610129e+00 | -3.593709e+00 | -3.621035e+00 | -3.912132e+00 | -3.924080e+00 | -3.670515e+00 | -1.952642e+00 | -3.565266e+00 | -2.146557e+00 |
| 25% | -6.701758e-01 | -6.670909e-01 | -6.835581e-01 | -6.800196e-01 | -6.682490e-01 | -6.670017e-01 | -9.264151e-01 | -6.859074e-01 | -6.698788e-01 | -6.765303e-01 | -6.758763e-01 | -6.712561e-01 | -6.837196e-01 | -6.817490e-01 | -6.791238e-01 | -6.738840e-01 | -6.684718e-01 | -9.895199e-01 | -6.780318e-01 | -8.897937e-01 |
| 50% | 2.088883e-03 | 4.061022e-03 | 6.491199e-04 | 3.440946e-03 | -3.559668e-03 | 6.739244e-03 | 2.121647e-03 | -1.018180e-02 | 8.129254e-04 | 1.343290e-02 | -6.378701e-03 | -1.497902e-03 | 1.165557e-03 | 9.576464e-03 | 8.139716e-03 | -6.609107e-03 | -8.836017e-03 | 5.153569e-02 | -2.638754e-03 | -1.884091e-01 |
| 75% | 6.776271e-01 | 6.633211e-01 | 6.685248e-01 | 6.733688e-01 | 6.826127e-01 | 6.669742e-01 | 6.198403e-01 | 6.767314e-01 | 6.737767e-01 | 6.555822e-01 | 6.713280e-01 | 6.741036e-01 | 6.808373e-01 | 6.771723e-01 | 6.740690e-01 | 6.795435e-01 | 6.981673e-01 | 5.261333e-01 | 6.744299e-01 | 8.889074e-01 |
| max | 3.786736e+00 | 3.700328e+00 | 3.819858e+00 | 3.731669e+00 | 3.686054e+00 | 3.725068e+00 | 3.512268e+00 | 3.896716e+00 | 3.842502e+00 | 4.236235e+00 | 3.556807e+00 | 3.783354e+00 | 3.717151e+00 | 3.382842e+00 | 3.849388e+00 | 3.861931e+00 | 3.529508e+00 | 3.719336e+00 | 3.887281e+00 | 3.033482e+00 |
2. Feature Dimenstion Reduce
- using
PCA: one of represenative method known as Linear Dimension Reduction. PCA(Principal Component Analysis)
from sklearn.decomposition import PCA
pca = PCA(n_components=5)
reduced_dim = pca.fit_transform(data)
reduced_dim = pd.DataFrame(reduced_dim, columns=['principal_comp1','principal_comp2','principal_comp3','principal_comp4','principal_comp5'])
reduced_dim
| principal_comp1 | principal_comp2 | principal_comp3 | principal_comp4 | principal_comp5 | |
|---|---|---|---|---|---|
| 0 | -1.618868 | 0.604217 | 0.781736 | -0.547131 | -0.846699 |
| 1 | -1.720829 | -0.031994 | 0.435634 | -2.228988 | 0.958939 |
| 2 | -1.926017 | -1.115552 | 0.746879 | 2.104426 | -1.363396 |
| 3 | -1.350258 | 0.556883 | -0.693035 | -2.326694 | 1.298485 |
| 4 | -0.746567 | 0.096190 | -0.198689 | 1.189678 | -1.028490 |
| ... | ... | ... | ... | ... | ... |
| 9995 | 3.045021 | 0.325163 | -0.428845 | 1.365473 | -1.105546 |
| 9996 | -0.575423 | 0.545010 | 1.186840 | 0.204053 | -0.563458 |
| 9997 | 1.088324 | -0.578450 | 0.264246 | -0.760159 | -0.259600 |
| 9998 | -2.304876 | -0.420677 | -1.557204 | 1.736730 | 1.008181 |
| 9999 | -2.319175 | 0.200419 | 0.456647 | 0.820593 | -0.812472 |
10000 rows × 5 columns
3. Feature Character
1) Test Classifier using RandomForestClassifier
2) Charateristic: explained variance ratio summation is 34.703%, i.e 34.7% information of original data was remained
from sklearn.ensemble import RandomForestClassifier
# 1) Test Classifier using RandomForestClassifier
clf=RandomForestClassifier(n_estimators=90)
clf.fit(reduced_dim,y)
score = clf.score(reduced_dim, y)
print("RandomForestClassifier Score is {} %".format(score*100))
print("Dimension Reduction was executed well!")
RandomForestClassifier Score is 100.0 %
Dimension Reduction was executed well!
import numpy as np
# 2) Charateristic
print('Eigen_value :', pca.explained_variance_)
#Explained Variance Ratio는 주성분 벡터가 이루는 축에 투영(projection)한 결과의 분산의 비율을 말하며, 각 eigenvalue의 비율과 같은 의미
print('Explained variance ratio :', pca.explained_variance_ratio_)
print('Information Reduction Percet is {}%'.format(np.sum(pca.explained_variance_ratio_)*100))
df = pd.DataFrame(pca.explained_variance_ratio_, index=['principal_comp1','principal_comp2','principal_comp3','principal_comp4','principal_comp5'], columns=['explained_variance_ratio_'])
df.plot.pie(y='explained_variance_ratio_', figsize=(7,7), legend=False)
Eigen_value : [2.76572688 1.05425708 1.04845199 1.03346204 1.02878758]
Explained variance ratio : [0.13827252 0.05270758 0.05241736 0.05166793 0.05143424]
Information Reduction Percet is 34.64996247376067%
<AxesSubplot:ylabel='explained_variance_ratio_'>
plt.figure()
colors=['navy','red']
aug = pd.concat([reduced_dim, y],axis=1)
fig=plt.figure(figsize=(30,10))
ax1=fig.add_subplot(1,4,1)
ax2=fig.add_subplot(1,4,2)
ax3=fig.add_subplot(1,4,3)
ax4=fig.add_subplot(1,4,4)
for color, label in zip(colors, [0, 1]):
ax1.scatter(x=aug.loc[aug['Label'] == label]['principal_comp1'], y=aug.loc[aug['Label'] == label]['principal_comp2'], c=color, s=1, alpha=0.9, label=str(label))
ax1.set_xlabel('principal_comp1')
ax1.set_ylabel('principal_comp2')
for color, label in zip(colors, [0, 1]):
ax2.scatter(x=aug.loc[aug['Label'] == label]['principal_comp1'], y=aug.loc[aug['Label'] == label]['principal_comp3'], c=color, s=1, alpha=0.9)
ax2.set_xlabel('principal_comp1')
ax2.set_ylabel('principal_comp3')
for color, label in zip(colors, [0, 1]):
ax3.scatter(x=aug.loc[aug['Label'] == label]['principal_comp1'], y=aug.loc[aug['Label'] == label]['principal_comp4'], c=color, s=1, alpha=0.9)
ax3.set_xlabel('principal_comp1')
ax3.set_ylabel('principal_comp4')
for color, label in zip(colors, [0, 1]):
ax4.scatter(x=aug.loc[aug['Label'] == label]['principal_comp1'], y=aug.loc[aug['Label'] == label]['principal_comp5'], c=color, s=1, alpha=0.9)
ax4.set_xlabel('principal_comp1')
ax4.set_ylabel('principal_comp5')
<Figure size 432x288 with 0 Axes>
corrmat = aug.corr()
plt.figure(figsize=(10,10))
sns.heatmap(corrmat, annot=True)
<AxesSubplot:>
끝
sungwook
Mr