Dimensionality Reduction with t-SNE

In [5]:

    

%matplotlib inline
import matplotlib.pyplot as plt
import seaborn as sns
import pandas as pd

from sklearn import manifold, datasets
from matplotlib.colors import ListedColormap


    

In [6]:

    

iris = datasets.load_iris()
tsne = manifold.TSNE(n_components=2, learning_rate=100)
new_dim = tsne.fit_transform(iris.data)


    

In [7]:

    

df = pd.DataFrame(new_dim, columns=['X', 'Y'])
df['label'] = iris.target
df.head()


    

    
Out[7]:






  

    

      

      
X
      
Y
      
label
    
  
  

    

      
0
      
-5.585848
      
5.384402
      
0
    
    

      
1
      
-3.638064
      
6.461707
      
0
    
    

      
2
      
-4.018937
      
7.319159
      
0
    
    

      
3
      
-3.738372
      
7.314444
      
0
    
    

      
4
      
-5.892209
      
5.577005
      
0
    
  

In [8]:

    

fig = plt.figure()
fig.suptitle('t-SNE', fontsize=14, fontweight='bold')
ax = fig.add_subplot(111)

plt.scatter(df[df.label == 0].X, df[df.label == 0].Y, color='red', label=iris.target_names[0])
plt.scatter(df[df.label == 1].X, df[df.label == 1].Y, color='blue', label=iris.target_names[1])
plt.scatter(df[df.label == 2].X, df[df.label == 2].Y, color='green', label=iris.target_names[2])

plt.legend(bbox_to_anchor=(1.25, 1))


    

    
Out[8]:





<matplotlib.legend.Legend at 0x10c533da0>