-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathApep_HMC_HPC.py
269 lines (203 loc) · 9.67 KB
/
Apep_HMC_HPC.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
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
# -*- coding: utf-8 -*-
"""
Created on Sun Mar 17 13:05:12 2024
@author: ryanw
"""
import os
import multiprocessing
os.environ["XLA_FLAGS"] = "--xla_force_host_platform_device_count={}".format(
multiprocessing.cpu_count()
)
import numpy as np
import jax_tqdm as jtqdm
import jax.numpy as jnp
from jax import jit, vmap, grad
import jax.lax as lax
import jax.scipy.stats as stats
import matplotlib.pyplot as plt
import time
import pickle
import numpyro, jax
import numpyro.distributions as dists
from numpyro.infer.util import initialize_model
from glob import glob
from astropy.io import fits
import WR_Geom_Model as gm
import WR_binaries as wrb
# apep = wrb.apep.copy()
numpyro.enable_x64()
system = wrb.apep.copy()
### --- INFERENCE --- ###
# n = 256 # standard
n = 600 # VISIR
# n = 898 # JWST
@jit
def smooth_histogram2d(particles, weights, stardata):
im_size = n
x = particles[0, :]
y = particles[1, :]
xbound, ybound = jnp.max(jnp.abs(x)), jnp.max(jnp.abs(y))
bound = jnp.max(jnp.array([xbound, ybound])) * (1. + 2. / im_size)
xedges, yedges = jnp.linspace(-bound, bound, im_size+1), jnp.linspace(-bound, bound, im_size+1)
return gm.smooth_histogram2d_base(particles, weights, stardata, xedges, yedges, im_size)
@jit
def smooth_histogram2d_w_bins(particles, weights, stardata, xbins, ybins):
im_size = n
return gm.smooth_histogram2d_base(particles, weights, stardata, xbins, ybins, im_size)
pscale = 1000 * 23/512 # mas/pixel, (Yinuo's email said 45mas/px, but I think the FOV is 23x23 arcsec for a 512x512 image?)
vlt_years = [2016, 2017, 2018, 2024]
vlt_data = {}
flattened_vlt_data = {}
directory = "Data/VLT"
fnames = glob(directory + "/*.fits")
for i, fname in enumerate(fnames):
data = fits.open(fname)[0].data
length = data.shape[0]
X = jnp.linspace(-1., 1., length) * pscale * length/2 / 1000
Y = X.copy()
xs, ys = jnp.meshgrid(X, Y)
data = jnp.array(data)
data = data - jnp.percentile(data, 84)
data = data/jnp.max(data)
data = jnp.maximum(data, 0)
data = jnp.abs(data)**0.5
vlt_data[vlt_years[i]] = data
flattened_vlt_data[vlt_years[i]] = data.flatten()
# big_flattened_data = jnp.concatenate([flattened_vlt_data[year] for year in vlt_years])
xbins = X
ybins = Y
print("Data loaded in well.")
system_params = system.copy()
def apep_model():
params = system_params.copy()
# m1 = numpyro.sample("m1", dists.Normal(apep['m1'], 5.))
# m2 = numpyro.sample("m2", dists.Normal(apep['m2'], 5.))
params['eccentricity'] = numpyro.sample("eccentricity", dists.Uniform(0.4, 0.95))
# params['eccentricity'] = numpyro.sample("eccentricity", dists.Normal(apep['eccentricity'], 0.05, support=dists.constraints.interval(0., 0.95)))
params['inclination'] = numpyro.sample("inclination", dists.Uniform(0., 50.))
params["asc_node"] = numpyro.sample("asc_node", dists.Uniform(100., 210.))
params["arg_peri"] = numpyro.sample("arg_peri", dists.Uniform(-30., 50.))
# open_angle = numpyro.sample("open_angle", dists.Normal(apep['open_angle'], 10.))
params['open_angle'] = numpyro.sample("open_angle", dists.Uniform(70, 140.))
# period = numpyro.sample("period", dists.Normal(apep['period'], 40.))
# distance = numpyro.sample("distance", dists.Normal(apep['distance'], 500.))
params["windspeed1"] = numpyro.sample("windspeed1", dists.Uniform(300., 1500.))
# windspeed2 = numpyro.sample("windspeed2", dists.Normal(apep['windspeed2'], 200.))
params['turn_on'] = numpyro.sample("turn_on", dists.Uniform(-150., -60.))
params['turn_off'] = numpyro.sample("turn_off", dists.Uniform(50., 179.))
# oblate = numpyro.sample("oblate", dists.Uniform(0., 1.))
# orb_sd = numpyro.sample("orb_sd", dists.Exponential(1./10.))
# orb_amp = numpyro.sample("orb_amp", dists.Exponential(1./0.1))
# orb_min = numpyro.sample("orb_min", dists.Uniform(0., 360.))
# az_sd = numpyro.sample("az_sd", dists.Exponential(1./10.))
# az_amp = numpyro.sample("az_amp", dists.Exponential(1./0.1))
# az_min = numpyro.sample("az_min", dists.Uniform(0., 360.))
# comp_incl = numpyro.sample('comp_incl', dists.Normal(apep['comp_incl'], 10))
# comp_az = numpyro.sample('comp_az', dists.Normal(apep['comp_az'], 10))
# comp_open = numpyro.sample("comp_open", dists.Normal(apep['comp_open'], 10.))
# comp_reduction = numpyro.sample("comp_reduction", dists.Uniform(0., 2.))
# comp_plume = numpyro.sample("comp_plume", dists.Uniform(0., 2.))
# phase = numpyro.sample("phase", dists.Uniform(0., 1.))
params['phase'] = numpyro.sample("phase", dists.Uniform(0., 0.99))
params['sigma'] = numpyro.sample("sigma", dists.Uniform(1., 10.))
# err_lum_factor = numpyro.sample("err_a", dists.LogUniform(1e-1, 4))
# err_const = numpyro.sample("err_b", dists.LogUniform(1e-4, 1))
err_lum_factor = numpyro.sample("err_a", dists.Uniform(1e-1, 5.))
# err_const = numpyro.sample("err_b", dists.Uniform(0., 1.))
err_const = 0.05
# for year in vlt_years:
# year_params = params.copy()
# year_params['phase'] -= (2024 - year) / params['period']
# samp_particles, samp_weights = gm.dust_plume(year_params)
# # # offset params to ensure the model lines up with the image:
# # offset_x = numpyro.sample(f"offset_x_{year}", dists.Uniform(-6., 6.))
# # offset_y = numpyro.sample(f"offset_y_{year}", dists.Uniform(-6., 6.))
# # samp_particles = samp_particles.at[0, :].add(offset_x)
# # samp_particles = samp_particles.at[1, :].add(offset_y)
# _, _, samp_H = smooth_histogram2d_w_bins(samp_particles, samp_weights, year_params, xbins, ybins)
# samp_H = gm.add_stars(xbins, ybins, samp_H, year_params)
# # samp_H.at[280:320, 280:320].set(0.)
# samp_H = samp_H.flatten()
# # samp_H = jnp.nan_to_num(samp_H, 1e4)
# data = flattened_vlt_data[year]
# # now we need to deal with the error in the images:
# # err_lum_factor = numpyro.sample(f"err_a_{year}", dists.LogUniform(1e-3, 4))
# # err_const = numpyro.sample(f"err_b_{year}", dists.LogUniform(1e-6, 1))
# err = jnp.sqrt(err_lum_factor * data + err_const**2)
# # err = 0.2
# with numpyro.plate('plate', len(data)):
# numpyro.sample(f'obs_{year}', dists.Normal(samp_H, err), obs=data)
year = 2016
year_params = params.copy()
year_params['phase'] -= (2024 - year) / params['period']
samp_particles, samp_weights = gm.dust_plume(year_params)
# # offset params to ensure the model lines up with the image:
# offset_x = numpyro.sample(f"offset_x_{year}", dists.Uniform(-6., 6.))
# offset_y = numpyro.sample(f"offset_y_{year}", dists.Uniform(-6., 6.))
# samp_particles = samp_particles.at[0, :].add(offset_x)
# samp_particles = samp_particles.at[1, :].add(offset_y)
_, _, samp_H = smooth_histogram2d_w_bins(samp_particles, samp_weights, year_params, xbins, ybins)
# samp_H = gm.add_stars(xbins, ybins, samp_H, year_params)
# samp_H.at[280:320, 280:320].set(0.)
samp_H = samp_H.flatten()
# samp_H = jnp.nan_to_num(samp_H, 1e4)
data = flattened_vlt_data[year]
err = jnp.sqrt(err_lum_factor * data + err_const**2)
with numpyro.plate('plate', len(data)):
numpyro.sample('obs', dists.Normal(samp_H, err), obs=data)
rand_time = int(time.time() * 1e8)
# rng_key = jax.random.key(rand_time)
rng_key = jax.random.PRNGKey(rand_time)
init_val = wrb.apep.copy()
for year in vlt_years:
init_val[f"err_a_{year}"] = 1.
init_val[f"err_b_{year}"] = 1e-2
init_val[f"offset_x_{year}"] = 0.
init_val[f"offset_y_{year}"] = 0.
init_val['err_a'] = 2.
init_val['err_b'] = 1e-1
init_val['sigma'] = 4.
# num_chains = min(10, len(jax.devices()))
num_chains = 1
print("Num Chains = ", num_chains)
sampler = numpyro.infer.MCMC(numpyro.infer.NUTS(apep_model,
max_tree_depth=5,
init_strategy=numpyro.infer.initialization.init_to_value(values=init_val)
),
num_chains=num_chains,
num_samples=1000,
num_warmup=200,
progress_bar=True)
t1 = time.time()
print("Running HMC Now.")
sampler.warmup(rng_key, collect_warmup=True)
# sampler.run(rng_key)
print("HMC Finished successfully.")
t2 = time.time()
print("Time taken = ", t2 - t1)
results = sampler.get_samples(group_by_chain=True)
results_flat = sampler.get_samples()
run_num = 2
with open(f'HPC/run_{run_num}/{rand_time}', 'wb') as file:
pickle.dump(results, file)
with open(f'HPC/run_{run_num}/{rand_time}_flat', 'wb') as file:
pickle.dump(results_flat, file)
with open(f'HPC/run_{run_num}/{rand_time}_last_state', 'wb') as file:
pickle.dump(sampler.last_state, file)
with open(f'HPC/run_{run_num}/{rand_time}_sampler', 'wb') as file:
pickle.dump(sampler, file)
import corner
corner.corner(results)
ndim = len(results.keys())
params = list(results.keys())
fig, axes = plt.subplots(figsize=(7, 2 * ndim), nrows=ndim, sharex=True, gridspec_kw={'hspace':0})
for i in range(ndim):
param_vals = results[params[i]]
if len(param_vals.shape) > 1:
for j in range(param_vals.shape[0]):
axes[i].scatter(np.arange(len(param_vals[j, :])), param_vals[j, :], s=1, rasterized=True)
else:
axes[i].scatter(np.arange(len(param_vals)), param_vals, s=1, rasterized=True)
# axes[i].set(ylabel=param_labels[params[i]])
axes[i].set(ylabel=params[i])
axes[-1].set(xlabel='Walker Iteration')