-
Notifications
You must be signed in to change notification settings - Fork 0
/
Fig3analysis.py
172 lines (127 loc) · 5.46 KB
/
Fig3analysis.py
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
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
import numpy as np
import matplotlib as mpl
mpl.use('Agg')
import matplotlib.pyplot as plt
def read_csv(f,header = False,skip = 1):
F = open(f,"r")
out = []
n = -1
for l in F:
n += 1
if n%skip == 0:
temp = l.split(",")
if header and n == 0:
continue#out.append([t.split("\n")[0] for t in temp])
else:
out.append([float(t.split("\n")[0]) for t in temp])
out = np.array(out)[:1000*int(len(out)/1000)]
print(out.shape)
return out
test_data = np.loadtxt("./FF_test_curves.csv")
model_output = np.reshape(read_csv("./tuning_curveswgan_FF_40_40_new.csv",header=True),[-1,1000,27])
print(test_data.shape)
print(model_output.shape)
from scipy.stats import ks_2samp as KS
from sklearn import linear_model
#things to define: r-squared, complexity, sparsity
def sparsity(tc):
l1 = len([t for t in tc if t < 5])
return float(l1)/len(tc)
def rsquared(tc,getp = False):
tcrs = np.copy(np.reshape(tc,[3,3,3,1]))
coord = np.array([[[[x,y,z] for x in range(3)] for y in range(3)] for z in range(3)])
comb = np.concatenate([coord,tcrs],axis = -1)
tcf = np.reshape(comb,[-1,4])
LF = linear_model.LinearRegression()
LF.fit(tcf[:,:3],tcf[:,-1])
rs = LF.score(tcf[:,:3],tcf[:,-1])
if getp:
proj = [[np.dot(LF.coef_,t[:3]),t[-1]] for t in tcf]
proj.sort(key = lambda x:x[0])
proj = np.array(proj)
proj[:,0] -= np.min(proj[:,0])
proj[:,0] /= (np.max(proj[:,0]) - np.min(proj[:,0]) + .001)
return np.copy(proj)
return rs
def com(tc):
tcrs = np.copy(np.reshape(tc,[3,3,3,1]))
tcrs = (tcrs - tcrs.mean())/(tcrs.max() - tcrs.min())
dif = np.stack([np.ravel(tcrs[1:]-tcrs[:-1]),np.ravel(tcrs[:,1:]-tcrs[:,:-1]),np.ravel(tcrs[:,:,1:]-tcrs[:,:,:-1])])
return np.std(np.abs(dif))
sk = 100
def good(t):
a = (np.max(t)>5)
b = ((np.max(t) - np.min(t)) > .01)
return (a and b)
test_sp = [sparsity(t) for t in test_data]
model_sp = [[sparsity(t) for t in m] for m in model_output[::sk]]
KSsp = [KS(test_sp,m)[0] for m in model_sp]
test_RS = [rsquared(t) for t in test_data if good(t)]
model_RS = [np.array([rsquared(t) for t in m if good(t)]) for m in model_output[::sk]]
test_PR = np.array([rsquared(t,getp = True) for t in test_data if good(t)])
model_PR = np.array([np.array([rsquared(t,getp = True) for t in m if good(t)]) for m in model_output[::sk]])
KSrs = [KS(test_RS,m)[0] for m in model_RS]
test_C = [com(t) for t in test_data if good(t)]
model_C = [np.array([com(t) for t in m if good(t)]) for m in model_output[::sk]]
KSC = [KS(test_C,m)[0] for m in model_C]
test_R = [np.mean(r) for r in test_data]
model_R = [np.array([np.mean(r) for r in m]) for m in model_output[::sk]]
KSR = [KS(test_R,m)[0] for m in model_R]
plt.clf()
fig = plt.figure(figsize = (5,6))
#ax = fig.add_axes([0,0,1,1])
ax1 = fig.add_axes([.15,.08,.78,.15])
ax2 = fig.add_axes([.15,.32,.33,.15])
ax3 = fig.add_axes([.6,.32,.33,.15])
ax4 = fig.add_axes([.15,.55,.33,.15])
ax5 = fig.add_axes([.6,.55,.33,.15])
ax6 = fig.add_axes([.15,.8,.33,.15])
ax7 = fig.add_axes([.6,.8,.33,.15])
gs = [500 * s * sk for s in range(len(KSR))]
ax1.plot(gs,KSR,label = "Rate")
ax1.plot(gs,KSsp,label = "Sparsity")
ax1.plot(gs,KSC,label = "Complexity")
ax1.plot(gs,KSrs,label = "R-Squared")
ax1.set_xlabel("Generator Training Step")
ax1.set_ylabel("KS Distance")
ax1.set_ylim([0,1])
ax1.legend(loc = [.4,.4],fontsize = 8)
norm = True
nbin = 25
ax2.hist(test_C,np.linspace(0,.5,nbin),histtype = "step",normed = norm,color = "k",linewidth = 2,label = "truth")
ax2.hist(model_C[0],np.linspace(0,.5,nbin),histtype = "step",normed = norm,color = "c",linewidth = 2,label = "init")
ax2.hist(model_C[-1],np.linspace(0,.5,nbin),histtype = "step",normed = norm,color = "r",linewidth = 2,label = "final")
ax2.set_xlabel("Complexity")
ax2.set_ylabel("Probability Mass")
ax2.set_ylim([0,15])
ax2.legend()
ax3.hist(test_RS,np.linspace(0,1,nbin),histtype = "step",normed = norm,color = "k",linewidth = 2)
ax3.hist(model_RS[0],np.linspace(0,1,nbin),histtype = "step",normed = norm,color = "c",linewidth = 2)
ax3.hist(model_RS[-1],np.linspace(0,1,nbin),histtype = "step",normed = norm,color = "r",linewidth = 2)
ax3.set_xlabel("R-Squared")
ax3.set_ylabel("Probability Mass")
ax3.set_ylim([0,5])
ax4.hist(test_R,np.linspace(0,50,nbin),histtype = "step",normed = norm,color = "k",linewidth = 2)
ax4.hist(model_R[0],np.linspace(0,50,nbin),histtype = "step",normed = norm,color = "c",linewidth = 2)
ax4.hist(model_R[-1],np.linspace(0,50,nbin),histtype = "step",normed = norm,color = "r",linewidth = 2)
ax4.set_xlabel("Rate")
ax4.set_ylabel("Probability Mass")
ax4.set_ylim([0,.1])
ax5.hist(test_sp,np.linspace(0,1,10),histtype = "step",normed = norm,color = "k",linewidth = 2)
ax5.hist(model_sp[0],np.linspace(0,1,10),histtype = "step",normed = norm,color = "c",linewidth = 2)
ax5.hist(model_sp[-1],np.linspace(0,1,10),histtype = "step",normed = norm,color = "r",linewidth = 2)
ax5.set_xlabel("Sparsity")
ax5.set_ylabel("Probability Mass")
ax5.set_ylim([0,5])
for k in range(5):
ax7.plot(test_PR[k,:,0],test_PR[k,:,1],"k")
ax7.plot(model_PR[-1][k,:,0],model_PR[-1][k,:,1],"r")
ax7.set_xlabel("PP location")
ax7.set_ylabel("Rate")
for k in range(5):
ax6.plot(test_PR[k,:,0],test_PR[k,:,1],"k")
ax6.plot(model_PR[0][k,:,0],model_PR[0][k,:,1],"r")
ax6.set_xlabel("PP location")
ax6.set_ylabel("Rate")
#plt.show()
plt.savefig("./Fig3_pythonversion.pdf")