TensorDecomp/tests.py at master · adamjermyn/TensorDecomp · 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
import numpy as np
import brute
import greedy
import greedy_variable

import matplotlib.pyplot as plt
plt.style.use('ggplot')

## Random examples

print('Random Normal:')
print('---')

dataB = []
dataG = []
dataG2 = []

for n in range(4, 10):

	dB = []
	dG = []
	dG2 = []

	for _ in range(100):
		x = np.random.randn(*[2 for _ in range(n)])

		b = brute.findBest(x, 1e-6)
		g = greedy.findBest(x, 1e-6)
		g2 = greedy_variable.findBest(x, 1e-6)


		dB.append(sum([v.size for v in b[2]]) * 1./x.size)
		dG.append(sum([v.size for v in g]) * 1./x.size)
		dG2.append(sum([v.size for v in g2]) * 1./x.size)

	dataB.append(np.average(dB))
	dataG.append(np.average(dG))
	dataG2.append(np.average(dG2))

for n in range(10,16):
	dG = []
	dG2 = []

	for _ in range(100):
		x = np.random.randn(*[2 for _ in range(n)])

		g = greedy.findBest(x, 1e-6)
		dG.append(sum([v.size for v in g]) * 1./x.size)

		g2 = greedy_variable.findBest(x, 1e-6)
		dG2.append(sum([v.size for v in g2]) * 1./x.size)

	dataG.append(np.average(dG))
	dataG2.append(np.average(dG2))


import matplotlib.pyplot as plt
fig = plt.figure(figsize=(5,4))
ax = plt.subplot(111)
plt.plot(range(4,10), dataB, label='Brute force')
plt.plot(range(4,16), dataG, label='Greedy')
plt.plot(range(4,16), dataG2, label='Greedy (Variable)')
plt.xlabel('Number of Indices')
plt.ylabel('Compression Ratio')
plt.legend()
plt.tight_layout()
plt.savefig('../normal.pdf')