-
Notifications
You must be signed in to change notification settings - Fork 34
/
bernoulli_slds.py
167 lines (138 loc) · 5.3 KB
/
bernoulli_slds.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
from __future__ import division
import numpy as np
import numpy.random as npr
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
from pybasicbayes.distributions import Regression, Gaussian
from pybasicbayes.util.text import progprint_xrange
from pypolyagamma.distributions import BernoulliRegression
from pyslds.models import HMMCountSLDS, WeakLimitStickyHDPHMMCountSLDS
npr.seed(0)
cmap = "jet"
### Hyperparameters
K, Kmax, D_obs, D_latent = 2, 10, 10, 2
mu_init = np.zeros(D_latent)
mu_init[0] = 1.0
sigma_init = 0.01 * np.eye(D_latent)
# Create an SLDS with stable dynamics matrices
def random_rotation(n,theta):
rot = np.array([[np.cos(theta), -np.sin(theta)],
[np.sin(theta), np.cos(theta)]])
out = np.zeros((n,n))
out[:2,:2] = rot
q = np.linalg.qr(np.random.randn(n,n))[0]
return q.dot(out).dot(q.T)
As = [random_rotation(D_latent, np.pi/24.),
random_rotation(D_latent, np.pi/8.)]
# Start with a random emission matrix
C = np.random.randn(D_obs, D_latent)
b = -2.0 * np.ones((D_obs, 1))
init_dynamics_distns = [Gaussian(mu=mu_init, sigma=sigma_init) for _ in range(K)]
dynamics_distns = [Regression(A=A, sigma=0.01*np.eye(D_latent)) for A in As]
emission_distns = BernoulliRegression(D_obs, D_latent, A=C, b=b)
truemodel = HMMCountSLDS(
dynamics_distns=dynamics_distns,
emission_distns=emission_distns,
init_dynamics_distns=init_dynamics_distns,
alpha=3., init_state_distn='uniform')
### Generate data from an SLDS
# Manually create the states object with the mask
T = 1000
stateseq = np.repeat(np.arange(T//100) % 2, 100).astype(np.int32)
statesobj = truemodel._states_class(model=truemodel, T=stateseq.size, stateseq=stateseq)
statesobj.generate_gaussian_states()
data = statesobj.data = statesobj.generate_obs()
# Manually mask off chunks of data
mask = np.ones_like(data, dtype=bool)
chunksz = 50
for i,offset in enumerate(range(0,T,chunksz)):
j = i % (D_obs + 1)
if j < D_obs:
mask[offset:min(offset+chunksz, T), j] = False
if j == D_obs:
mask[offset:min(offset+chunksz, T), :] = False
statesobj.mask = mask
truemodel.states_list.append(statesobj)
### Make a model
model = WeakLimitStickyHDPHMMCountSLDS(
init_dynamics_distns=
[Gaussian(
nu_0=5, sigma_0=3.*np.eye(D_latent),
mu_0=np.zeros(D_latent), kappa_0=0.01,
mu=mu_init, sigma=sigma_init
) for _ in range(Kmax)],
dynamics_distns=
[Regression(
A=np.eye(D_latent), sigma=np.eye(D_latent),
nu_0=D_latent+3,
S_0=D_latent*np.eye(D_latent),
M_0=np.zeros((D_latent, D_latent)),
K_0=D_latent*np.eye(D_latent),
) for _ in range(Kmax)],
emission_distns=BernoulliRegression(D_obs, D_latent),
alpha=3., gamma=3.0, kappa=100., init_state_distn='uniform')
model.add_data(data=data, mask=mask)
### Run a Gibbs sampler
N_samples = 500
def gibbs_update(model):
model.resample_model()
smoothed_obs = model.states_list[0].smooth()
return model.log_likelihood(), model.stateseqs[0], smoothed_obs
lls, z_smpls, smoothed_obss = \
zip(*[gibbs_update(model) for _ in progprint_xrange(N_samples)])
### Plot the log likelihood over iterations
plt.figure(figsize=(10,6))
plt.plot(lls,'-b')
plt.plot([0,N_samples], truemodel.log_likelihood() * np.ones(2), '-k')
plt.xlabel('iteration')
plt.ylabel('log likelihood')
### Plot the smoothed observations
fig = plt.figure(figsize=(10,10))
N_subplots = min(D_obs,6)
gs = gridspec.GridSpec(N_subplots*2+1,1)
zax = fig.add_subplot(gs[0])
zax.imshow(truemodel.stateseqs[0][None,:], aspect='auto', cmap=cmap)
zax.set_ylabel("Discrete \nstate", labelpad=20, multialignment="center", rotation=90)
zax.set_xticklabels([])
zax.set_yticks([])
given_data = data.copy()
given_data[~mask] = np.nan
masked_data = data.copy()
masked_data[mask] = np.nan
ylims = (-1.1*abs(data).max(), 1.1*abs(data).max())
xlims = (0, min(T,1000))
n_to_plot = np.arange(min(5, D_obs))
for i,j in enumerate(n_to_plot):
ax = fig.add_subplot(gs[1+2*i:1+2*(i+1)])
# Plot spike counts
given_ts = np.where(given_data[:,j]==1)[0]
ax.plot(given_ts, np.ones_like(given_ts), 'ko', markersize=5)
masked_ts = np.where(masked_data[:,j]==1)[0]
ax.plot(masked_ts, np.ones_like(masked_ts), 'o', markerfacecolor="gray", markeredgecolor="none", markersize=5)
# Plot the inferred rate
ax.plot([0], [0], 'b', lw=2, label="smoothed obs.")
ax.plot(smoothed_obss[-1][:,j], 'r', lw=2, label="smoothed pr.")
# Overlay the mask
ax.imshow(1-mask[:,j][None,:],cmap="Greys",alpha=0.25,extent=(0,T) + ylims, aspect="auto")
if i == 0:
plt.legend(loc="upper center", ncol=4, bbox_to_anchor=(0.5, 2.))
if i == N_subplots - 1:
plt.xlabel('time index')
ax.set_xlim(xlims)
ax.set_ylim(0, 1.1)
ax.set_ylabel("$x_%d(t)$" % (j+1))
### Plot the discrete state samples
fig = plt.figure(figsize=(8,4))
gs = gridspec.GridSpec(6,1)
ax1 = fig.add_subplot(gs[:-1])
ax2 = fig.add_subplot(gs[-1])
im = ax1.imshow(np.array(z_smpls), aspect='auto', interpolation="none", cmap=cmap)
ax1.autoscale(False)
ax1.set_ylabel("Iteration")
ax1.set_xticks([])
ax2.imshow(truemodel.stateseqs[0][None,:], aspect='auto', cmap=cmap)
ax2.set_ylabel("True", labelpad=27)
ax2.set_xlabel("Time")
ax2.set_yticks([])
fig.suptitle("Discrete state samples")
plt.show()