Embedding_visualization/Jointsne.py at master · GWUvision/Embedding_visualization · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
import numpy as np
#import torch
from sklearn import manifold
import os
#import torch
import Yoked_Tsne
import matplotlib.pyplot as plt


def draw(comEmbed,traLabel,valLabel,name,path,i):
    """ This function is used to draw joint tsne plot.

    Parameters
    ----------

    comEmbed : Tensor, shape (n_samples, n_features)
    Embedding representation of both training data and testting data.

    traLabel : List, length (n_tra_feafures)
    Label for each training instance.

    valLabel : List, length (n_tra_feafures)
    Label for each testting instance.

    name : String,
    Title for this joint tsne plot.

    path : String,
    Path to save this plot."""

    trasize = len(traLabel)
    a3=np.append(traLabel,valLabel)%7
    b3 = np.append(traLabel,valLabel)%12
    markers = ["." , "," , "o" , "v" , "^" , "<", ">","x","1","2","3","4"]
    colors = ['r','g','b','c','m', 'y', 'k']

    colorslist3=[]
    markerslist3 = []
    for item in a3:
        colorslist3.append(colors[item])
    for item in b3:
        markerslist3.append(markers[item])

    plt.figure(figsize=(90,30))
    plt.axis((-250,250,-250,250))
    plt.title(name)

    axe1=plt.subplot(1,3,1)
    axe1.set_title('tra')

    axe2=plt.subplot(1,3,2)
    axe2.set_title('All')

    axe3=plt.subplot(1,3,3)
    axe3.set_title('val')

    for k in range(0,trasize):
        axe1.scatter(comEmbed[k, 0], comEmbed[k, 1],c=colorslist3[k],marker=markerslist3[k])
        axe1.axis('on')
        axe2.scatter(comEmbed[k, 0], comEmbed[k, 1],c='r',marker=markerslist3[k],alpha=0.5)
        axe2.axis('on')
    for k in range(trasize,len(comEmbed)):
        axe2.scatter(comEmbed[k, 0], comEmbed[k, 1],c='b',marker=markerslist3[k],alpha=0.5)
        axe2.axis('on')
        axe3.scatter(comEmbed[k, 0], comEmbed[k, 1],c=colorslist3[k],marker=markerslist3[k])
        axe3.axis('on')

    plt.savefig(path+'Joint-sne_'+str(i)+'.jpg',dpi=300)
    plt.close()


def jointTsne(traPathList,valPathList,traLabelPath,valLabelPath,name=[''],path='./',alpha=8):

    """ This function is used to draw joint tsne plot.

    Parameters
    ----------

    traPath : String
    Path saving training embedding vectors

    valPath : String
    Path saving testing embedding vectors

    traLabelPath : String
    Path saving training label

    valLabelPath : String
    Path saving testing label

    name : String
    Title for this joint tsne plot.

    path : String
    Path to save this plot.

    alpha : int
    Hyperparameter for mapping cluster in different plot in together. Default:8

    """

    #traFvec = torch.load(traPathList[0])
    traFvec = np.load(traPathList[0])
    #valFvec = torch.load(valPathList[0])
    valFvec = np.load(valPathList[0])
    #traLabel = np.asarray(torch.load(traLabelPath))
    traLabel = np.load(traLabelPath)
    #valLabel = np.asarray(torch.load(valLabelPath))
    valLabel = np.load(valLabelPath)

    comFvec = np.concatenate((traFvec, valFvec), axis=0)
    comEmbed_0 = manifold.TSNE(n_components=2).fit_transform(comFvec)
    draw(comEmbed_0,traLabel,valLabel,name[0],path,'')

    for i in range(1,len(traPathList)):
        #traFvec = torch.load(traPathList[i])
        traFvec = np.load(traPathList[i])
        #valFvec = torch.load(valPathList[i])
        valFvec = np.load(valPathList[i])

        comFvec = np.concatenate((traFvec, valFvec), axis=0)
        comEmbed,_=Yoked_Tsne.Yoked_TSNE(n_components=2,init_Y=comEmbed_0).Yoke_transform(comFvec,0,alpha=10**-alpha,fixed_Y=True)
        draw(comEmbed,traLabel,valLabel,name,path,i)