-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdosa_vi.py
156 lines (132 loc) · 6.11 KB
/
dosa_vi.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
import pyro
import torch.distributions.constraints as constraints
import torch
import matplotlib.pyplot as plt
import pyro.distributions as dist
import pandas as pd
import numpy as np
import os
from pyro.optim import ClippedAdam
import seaborn as sns
os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
#model
def model(dosage,time, senescent=None):
B = pyro.sample("B", dist.Beta(10., 10.))
# b_0 = pyro.sample("b0", dist.Normal(0., 1.))
# b_1 = pyro.sample("b1", dist.Normal(0., 1.))
# b_2 = pyro.sample("b2", dist.Normal(0., 1.))
b_0 = pyro.sample("b0", dist.Beta(10., 10.))
b_1 = pyro.sample("b1", dist.Beta(10., 10.))
b_2 = pyro.sample("b2", dist.Beta(10., 10.))
mean= 100/(1+100*B*torch.exp(-(b_0/10+b_1/100*dosage+b_2/10*(dosage**2))*time))
with pyro.plate("data", len(time)):
return pyro.sample("obs", dist.Normal(mean, 1.), obs=senescent)
#pyro.render_model(model, model_args=(is_cont_africa, ruggedness, log_gdp), render_distributions=True)
def custom_guide(dosage,time, senescent=None):
B_loc = pyro.param('B_loc', torch.tensor(10.),
constraint=constraints.positive)
B_scale = pyro.param('B_scale', torch.tensor(10.),
constraint=constraints.positive)
# weights_loc = pyro.param('weights_loc', torch.zeros(3))
# weights_scale = pyro.param('weights_scale',torch.ones(3),
# constraint=constraints.positive)
# B = pyro.sample("B", dist.Beta(B_loc, B_scale))
# b_0 = pyro.sample("b0", dist.Normal(weights_loc[0], weights_scale[0]))
# b_1 = pyro.sample("b1", dist.Normal(weights_loc[1], weights_scale[1]))
# b_2 = pyro.sample("b2", dist.Normal(weights_loc[2], weights_scale[2]))
weights_loc = pyro.param('weights_loc',lambda: 10*torch.ones(3),
constraint=constraints.positive)
weights_scale = pyro.param('weights_scale',lambda: 10*torch.ones(3),
constraint=constraints.positive)
B = pyro.sample("B", dist.Beta(B_loc, B_scale))
b_0 = pyro.sample("b0", dist.Beta(weights_loc[0], weights_scale[0]))
b_1 = pyro.sample("b1", dist.Beta(weights_loc[1], weights_scale[1]))
b_2 = pyro.sample("b2", dist.Beta(weights_loc[2], weights_scale[2]))
return {B:"B","b_0": b_0, "b_1": b_1, "b_2": b_2}
#pyro.render_model(custom_guide, model_args=(is_cont_africa, ruggedness, log_gdp), render_params=True)
#d_trains=[0.1680, 0.5680, 0.9680, 1.3680, 1.7680, 2.1680, 2.5680, 2.9680]
d_trains=[0.1680,1.1680,3.1680,6.168,12.168,24.168,48.168, 96.168,192.168,384.168,768.168,1536.168,3072.168]
pred_sens_2=[]
qols_2=[]
pred_sens_5=[]
qols_5=[]
pred_sens_10=[]
qols_10=[]
sims=[]
datas=[]
for d_train in d_trains:
print("Dosage level: {}".format(d_train))
# load data
B, beta0, beta1, beta2 = 0.5, 0.05, 0.003, 0.01 # 0.5,0.1,0.03,0.08# ground-truth
t = np.arange(50)
r_d = lambda d: (beta0 + beta1 * d + beta2 * (d ** 2))
senescent = 100 / (1 + 100 * B * np.exp(-r_d(d_train) * t))
data = senescent + 1.* np.random.randn(50)
pyro.clear_param_store()
adam = ClippedAdam({"lr": 0.02})
elbo = pyro.infer.Trace_ELBO()
svi = pyro.infer.SVI(model, custom_guide, adam, elbo)
losses = []
time = torch.arange(50,dtype=torch.float )
data = torch.tensor(data, dtype=torch.float )
dosage = d_train*torch.ones(50,dtype=torch.float )
for step in range(2000): # Consider running for more steps.
loss = svi.step(dosage,time,data)
losses.append(loss)
if step % 100 == 0:
print("Elbo loss: {}".format(loss))
# plt.figure(figsize=(5, 2))
# plt.plot(losses)
# plt.xlabel("SVI step")
# plt.ylabel("ELBO loss")
# for name, value in pyro.get_param_store().items():
# print(name, pyro.param(name).data.cpu().numpy())
#
# # grab the learned variational parameters
# with pyro.plate("samples", 800, dim=-1):
# samples = custom_guide(dosage,time)
# fig = plt.figure(figsize=(10, 6))
# sns.histplot(samples['B'].detach().cpu().numpy(), kde=True, stat="density", label="African nations")
# plt.show()
predictive = pyro.infer.Predictive(model, guide=custom_guide, num_samples=800)
d_test=10*torch.ones(1,dtype=torch.float )
t_test=2*torch.ones(1,dtype=torch.float )
svi_samples = predictive(d_test,t_test)
svi_sens = svi_samples["obs"]
mean_qol=(-(svi_sens - svi_sens.mean(0)) ** 2).mean()
pred_sens_2.append(svi_sens)
qols_2.append(mean_qol)
d_test=10*torch.ones(1,dtype=torch.float )
t_test=5*torch.ones(1,dtype=torch.float )
svi_samples = predictive(d_test,t_test)
svi_sens = svi_samples["obs"]
mean_qol=(-(svi_sens - svi_sens.mean(0)) ** 2).mean()
pred_sens_5.append(svi_sens)
qols_5.append(mean_qol)
d_test = 10 * torch.ones(1, dtype=torch.float)
t_test = 10 * torch.ones(1, dtype=torch.float)
svi_samples = predictive(d_test, t_test)
svi_sens = svi_samples["obs"]
mean_qol = (-(svi_sens - svi_sens.mean(0)) ** 2).mean()
pred_sens_10.append(svi_sens)
qols_10.append(mean_qol)
sims.append(predictive(d_train,time)["obs"].mean(0))
datas.append(data)
pred_mean=torch.stack(pred_sens_2,dim=-1).squeeze().mean(0)
pred_std=torch.stack(pred_sens_2,dim=-1).squeeze().std(0)
plt.plot(torch.tensor(d_trains),pred_mean,color='blue')
plt.errorbar(torch.tensor(d_trains),pred_mean,pred_std,color='blue',alpha=0.2)
#plt.fill_between(torch.tensor(d_trains),pred_mean-pred_std,pred_mean+pred_std,alpha=0.2)
plt.xlabel("dosage levels/mGy ")
plt.ylabel("senescent/%")
plt.xscale('log')
pred_mean=torch.stack(qols_2)
plt.plot(torch.tensor(d_trains),pred_mean,color='blue')
#plt.fill_between(torch.tensor(d_trains),pred_mean-pred_std,pred_mean+pred_std,alpha=0.2)
plt.xlabel("dosage levels/mGy ")
plt.ylabel("Qol")
plt.xscale('log')
colors=['#1f77b4', '#ff7f0e', '#2ca02c', '#d62728', '#9467bd', '#8c564b', '#e377c2', '#7f7f7f', '#bcbd22', '#17becf']
for i,c in zip(sims,colors):
plt.plot(i,color=c)
plt.legend(d_trains)