num2vec/embedding.py at master · ayoungprogrammer/num2vec · 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
125
126
127
128
129
130
### t-Sne:
from rnn_seq_2 import SequenceModel
from sklearn import manifold
import matplotlib.pyplot as plt
import matplotlib.patheffects as PathEffects
import pandas as pd
import numpy as np

seq_model = SequenceModel()
seq_model.load()

print('Experiment')

xdata = [x for x in range(0,1000)]
squares = [x*x for x in range(1,40) if x*x < 1000]
pow2 = [2**x for x in range(1,11) if 2**x < 1000]


embedded_data = seq_model.embed_int(xdata)
tsne = manifold.TSNE(n_components=2, init='pca', random_state=42)

#import ipdb;ipdb.set_trace()

#normalized = (embedded_data - np.mean(embedded_data, axis=0)) / np.std(embedded_data, axis=0)
#X_tsne = tsne.fit_transform(normalized)

X_tsne1000 = tsne.fit_transform(embedded_data)
X_tsne100 = tsne.fit_transform(embedded_data[:100])

def scatter_tsne(X_tsne, nums, c):
    plt.scatter(X_tsne[nums][:,0], X_tsne[nums][:,1], c=c)

def plot_tsne(X_tsne, nums, fmt='o'):
    plt.plot(X_tsne[nums][:,0], X_tsne[nums][:,1], fmt)

def plot_label(pt, label, fc='red'):
    plt.annotate(label, pt, color=fc, path_effects=[PathEffects.withStroke(linewidth=2, foreground='w')])

def plot_num_embeddings():
    """
    Plot colors from 1 to 100
    """
    colors = [x%10 for x in range(100)]
    scatter_tsne(X_tsne100, xdata[:100], colors)
    for x in range(100):
        plot_label(X_tsne100[x], x)
    plt.savefig('img/100-embedding-ones.png')
    plt.clf()


    plot_tsne(X_tsne100, xdata[:100], '-o')
    plt.savefig('img/100-embedding-seq.png')
    plt.clf()

    """
    Plot colors from 1 to 1000
    """
    plot_tsne(X_tsne1000, xdata, '-o')
    plt.savefig('img/1000-embedding.png')
    plt.clf()

    """
    Plot colors by %
    """
    colors = [x % 10 for x in range(1000)]
    scatter_tsne(X_tsne1000, xdata, colors)
    plt.savefig('img/1000-embedding-last-digit.png')
    plt.clf()


with open('data/test-data.txt') as f:
    def split_row(row):
        cells = row.strip().split('\t')
        return cells[0], list(map(int, cells[1:]))
    test_data = list(map(split_row, f.readlines()))


seq_paths = []
seq_embedded = []
for line in test_data[:300]:
    vec = seq_model.embed_seq(line[1])
    seq_embedded.append(vec)
    seq_paths.append(line[0])


label_data = [
('x', [1, 2, 3, 4, 5, 6, 7, 8]),
('x+1', [2,3,4,5,6,7,8,9]),
('x+1334', [1335, 1336, 1337, 1338, 1339, 1340, 1341, 1342]),
('[10,3,4] repeat', [10,3,4,10,3,4,10,3]),
('[54,96,87, 23] repeat', [54, 96, 87, 23, 54, 96, 87, 23]),
('x^2', [1, 2, 4, 8, 16, 32, 64, 128]),
('3x^2', [3, 6, 12, 24, 48, 48, 192, 384]),
('100-x', [100, 99, 98, 97, 96, 95, 94, 93]),
('1000-x', [1000, 999, 998, 997, 996, 995, 994, 993]),
('8x', [1, 8, 16, 24, 32, 40, 48, 56]),
('500 + x', [501, 502, 503, 504, 505, 506, 507, 508]),
('601 - x', [600, 599, 598, 597, 596, 595, 594, 593]),
('9000 - 3x', [8997, 8994, 8991, 8988, 8985, 8982, 8979, 8976]),
('95234 - x', [95234, 95233, 95232, 95231, 95230, 95229, 95228, 95227]),
('51343 - 2x', [51343, 51341, 51339, 51337, 51335, 51333, 51331, 51329]),
('120000 + x', [120001, 120002, 120003, 120004, 120005, 120006, 120007, 120008]),
('215000 - x', [215499, 215498, 215497, 215496, 215495, 215494, 215493, 215493]),
]

label_embed = []
label_pat = []
for pat, seq in label_data:
    label_embed.append(seq_model.embed_seq(seq))
    label_pat.append(pat)


X_tsne_seq = tsne.fit_transform(np.matrix(seq_embedded + label_embed))
xdata = [x for x in range(00, len(X_tsne_seq))]

label_tsne_seq = X_tsne_seq[len(seq_embedded):]


plot_tsne(X_tsne_seq, xdata)
# for x in range(0, 300):
#    plot_label(X_tsne_seq[x], seq_paths[x], 'black')
for x in range(len(label_tsne_seq)):
    plot_label(X_tsne_seq[x+len(seq_embedded)], label_pat[x], 'red')
plt.savefig('img/tsne_seqs.png')

# for x in range(10):
#     plot_label(X_tsne, x)
#     plot_label(X_tsne, x*10)

plot_num_embeddings()