-
Notifications
You must be signed in to change notification settings - Fork 0
/
Ksumtest_moro_cardin.py
122 lines (103 loc) · 4.31 KB
/
Ksumtest_moro_cardin.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
# Regular Modules
import numpy as np
import matplotlib.pyplot as plt
import datetime
import scipy.integrate as scint
from numpy.random import default_rng
import numpy.ma as ma
import matplotlib.tri as tri
import scipy.special as sc
import scipy.sparse as sps
import datetime
import itertools
import time
# My Modules
import src.helpers as helpers
import src.model_systems as model_systems
import src.diffusion_map as dmap
def main():
# Setting inverse temperature for plotting
beta = 1.0
def potential(x): return model_systems.morocardin_potential(x)
xmin, xmax = -2, 2
ymin, ymax = -2, 2
#datasets = ['betainv3', 'metad', 'deltanet']
datasets = ['metad']
kernels = ['regular', 'mahalanobis']
eps_vals = 2.0**np.arange(-20, 10, 0.5)
Ksums = np.zeros((2, len(eps_vals)))
chi_logs = np.zeros_like(Ksums)
optimal_eps_vals = np.zeros(2)
for dataset in datasets:
fname = f"systems/MoroCardin/data/data_solution_{dataset}.npz"
inData = np.load(fname)
data = inData['data']
diffusion_list = inData['diffusion_list']
N = data.shape[1]
# Build Target Measure
target_beta = 1.
target_measure = np.zeros(N)
for i in range(N):
target_measure[i] = np.exp(-target_beta*potential(data[:, i]))
for i in range(2):
kernel = kernels[i]
[Ksums[i,:], chi_logs[i,:], optimal_eps_vals[i]] = Ksum_test(diffusion_list, eps_vals, data, kernel, target_measure)
fname = f"data/Ktest_{dataset}_{kernel}.npz"
#np.savez(fname, eps_vals=eps_vals, Ksum=Ksums[i,:], chi_log=chi_logs[i,:], errors=errors[i, :], optimal_eps = optimal_eps_vals[i])
print(f"optimal kde eps for {dataset}: {optimal_eps_vals[0]}")
print(f"optimal mahal eps for {dataset}: {optimal_eps_vals[1]}")
plt.figure()
plt.plot(eps_vals, Ksums[0,:])
plt.plot(eps_vals, Ksums[1,:])
plt.legend(['regular kernel', 'mahal kernel'])
plt.xscale("log", base=10)
plt.yscale("log", base=10)
plt.axvline(x=optimal_eps_vals[0], ls='--', c='C0')
plt.axvline(x=optimal_eps_vals[1], ls='--', c='C1')
fname = f"figures/logKsums_{dataset}"
#plt.savefig(fname, dpi=300)
plt.figure()
plt.plot(eps_vals, chi_logs[0, :])
plt.plot(eps_vals, chi_logs[1, :])
plt.legend(['regular kernel', 'mahal kernel'])
plt.title("log Ksums")
plt.xscale("log", base=10)
plt.yscale("log", base=10)
plt.title("dlog_Sum/dlog_eps")
fname = f"figures/dlogKsums_{dataset}"
plt.axvline(x=optimal_eps_vals[0], ls='--', c='C0')
plt.axvline(x=optimal_eps_vals[1], ls='--', c='C1')
#plt.savefig(fname, dpi=300)
plt.show()
def Ksum_test(diffusion_list, eps_vals, data, kernel, target_measure):
num_idx = eps_vals.shape[0]
Ksum = np.zeros(num_idx)
chi_log_analytical = np.zeros(num_idx)
for i in range(num_idx):
# Construct sparsified sqdists, kernel and generator with radius nearest neighbors
eps = eps_vals[i]
radius = None
# put a maximum for the radius in radius-nearest neighbors
if eps > 1:
radius = 3*np.sqrt(1)
if kernel == 'regular':
my_dmap = dmap.TargetMeasureDiffusionMap(epsilon=eps, pbc_dims=None, radius=radius, target_measure = target_measure)
elif kernel == 'mahalanobis':
my_dmap = dmap.TargetMeasureMahalanobisDiffusionMap(epsilon=eps, diffusion_list=diffusion_list, pbc_dims=None, radius=radius, target_measure = target_measure)
my_dmap.construct_generator(data)
Ksymm = my_dmap.get_kernel_reweight_symmetric()
sqdists = my_dmap.get_sqdists()
Ksum[i] = Ksymm.sum(axis=None)
# Compute deriv of log Ksum w.r.t log epsilon ('chi log')
if sps.issparse(sqdists) and sps.issparse(Ksymm):
mat = Ksymm.multiply(sqdists)
else:
mat = sqdists*Ksymm
chi_log_analytical[i] = mat.sum(axis=None) / ((2*eps)*Ksum[i])
print(f"epsilon: {eps}")
print(f"chi log: {chi_log_analytical[i]}")
print("\n")
optimal_eps = eps_vals[np.nanargmax(chi_log_analytical)]
return [Ksum, chi_log_analytical, optimal_eps]
if __name__ == "__main__":
main()