機械学習実践　次元削除と可視化

• 特徴量の作成
• カテゴリカルデータの前処理

などを行ったので再度次元削減を行い、結果を確認してみます

PCAでデータを2次元に圧縮して可視化

pca2 = PCA(n_components=2)
normalizeData = StandardScaler().fit_transform(fnData.values)
dim2PcaData = pca2.fit_transform(normalizeData)  
dim2PcaData = pd.DataFrame(dim2PcaData)  
dim2PcaData

0	1
0	-2.617890	1.230049
1	-0.883489	1.215867
2	-0.986500	-0.201632
3	-0.295452	-0.463923
4	-0.358544	-0.585817
…	…	…
995	-1.110524	-0.645389
996	0.584318	0.394180
997	-4.441661	-0.175817
998	-1.463922	0.208419
999	2.055026	0.118375

num = len(data['default.payment.next.month'].unique())
fig = plt.figure()
for i in range(num):
    target = dim2PcaData[data['default.payment.next.month'] == i]
    plt.scatter(x=target.iloc[:, 0], y=target.iloc[:, 1], label=str(i), alpha=0.5)

t-SNEでデータを2次元に圧縮して可視化

dim2TsneData = TSNE(n_components=2).fit_transform(fnData.iloc[:,:-1])
dim2TsneData = pd.DataFrame(dim2TsneData)
dim2TsneData

	0	1
0	30.335127	-39.175529
1	-9.754200	-14.677676
2	2.735364	-14.194139
3	17.963379	-5.905255
4	15.242915	-26.766068
…	…	…
995	-20.718533	-0.357178
996	-6.142908	0.424000
997	-24.534489	26.326624
998	5.997464	-32.502254
999	5.874343	14.879034

num = len(data['default.payment.next.month'].unique())
fig = plt.figure()
for i in range(num):
    target = dim2TsneData[data['default.payment.next.month'] == i]
    plt.scatter(x=target.iloc[:, 0], y=target.iloc[:, 1], label=str(i), alpha=0.5)

PCAでデータ3次元に圧縮して可視化

pca3 = PCA(n_components=3)
normalizeData = StandardScaler().fit_transform(fnData.iloc[:,:-1].values)
dim3PcaData = pca3.fit_transform(normalizeData)  
dim3PcaData = pd.DataFrame(dim3PcaData)  
dim3PcaData

0	1	2
0	-2.860972	0.440091	1.398824
1	-1.145599	0.763652	1.888266
2	-0.877400	-0.210833	-0.358869
3	-0.146323	-0.327251	-0.425060
4	-0.185465	-0.456240	-0.463068
…	…	…	…
995	-0.910446	-0.634648	-0.800143
996	0.464402	0.297923	1.341869
997	-4.330302	-1.001982	-0.917260
998	-1.444907	0.012196	0.109259
999	2.002282	0.451274	1.098129

num= len(data['default.payment.next.month'].unique())
fig = plt.figure()
ax = fig.gca(projection='3d')
for i in range(num):
    target =dim3PcaData[data['default.payment.next.month'] == i]
    ax.scatter(
        xs=target.iloc[:, 0], ys=target.iloc[:, 1],  zs=target.iloc[:, 2],label=str(i), alpha=0.5)

t-SNEでデータを3次元に圧縮して可視化

dim3TsneData = TSNE(n_components=3).fit_transform(fnData.iloc[:,:-1])
dim3TsneData = pd.DataFrame(dim3TsneData)  

num = len(data['default.payment.next.month'].unique())
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
for i in range(num):
    target = dim3TsneData[data['default.payment.next.month'] == i]
    ax.scatter(xs=target.iloc[:, 0], ys=target.iloc[:, 1],  zs=target.iloc[:, 2],label=str(i), alpha=0.5)

import plotly.graph_objs as go
from plotly.offline import init_notebook_mode, plot, iplot

init_notebook_mode(connected=True)