/usr/lib/python2.7/dist-packages/pymc/MCMC.py is in python-pymc 2.2+ds-1.1.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 | """
Class MCMC, which fits probability models using Markov Chain Monte Carlo, is defined here.
"""
__all__ = ['MCMC']
from .Model import Sampler
from .StepMethods import StepMethodRegistry, assign_method, DrawFromPrior
from .distributions import absolute_loss, squared_loss, chi_square_loss
import sys, time, pdb
import numpy as np
from .utils import crawl_dataless
from .six import print_
# from .progressbar import ProgressBar, Percentage, Bar, ETA, Iterations
from progressbar import ProgressBar
GuiInterrupt = 'Computation halt'
Paused = 'Computation paused'
class MCMC(Sampler):
"""
This class fits probability models using Markov Chain Monte Carlo. Each stochastic variable
is assigned a StepMethod object, which makes it take a single MCMC step conditional on the
rest of the model. These step methods are called in turn.
>>> A = MCMC(input, db, verbose=0)
:Parameters:
- input : module, list, tuple, dictionary, set, object or nothing.
Model definition, in terms of Stochastics, Deterministics, Potentials and Containers.
If nothing, all nodes are collected from the base namespace.
- db : string
The name of the database backend that will store the values
of the stochastics and deterministics sampled during the MCMC loop.
- verbose : integer
Level of output verbosity: 0=none, 1=low, 2=medium, 3=high
Inherits all methods and attributes from Model. Subclasses must define the _loop method:
- _loop(self, *args, **kwargs): Can be called after a sampling run is interrupted
(by pausing, halting or a KeyboardInterrupt) to continue the sampling run.
_loop must be able to handle KeyboardInterrupts gracefully, and should monitor
the sampler's status periodically. Available status values are:
- 'ready': Ready to sample.
- 'paused': A pause has been requested, or the sampler is paused. _loop should return control
as soon as it is safe to do so.
- 'halt': A halt has been requested, or the sampler is stopped. _loop should call halt as soon
as it is safe to do so.
- 'running': Sampling is in progress.
:SeeAlso: Model, Sampler, StepMethod.
"""
def __init__(self, input=None, db='ram', name='MCMC', calc_deviance=True, **kwds):
"""Initialize an MCMC instance.
:Parameters:
- input : module, list, tuple, dictionary, set, object or nothing.
Model definition, in terms of Stochastics, Deterministics, Potentials and Containers.
If nothing, all nodes are collected from the base namespace.
- db : string
The name of the database backend that will store the values
of the stochastics and deterministics sampled during the MCMC loop.
- verbose : integer
Level of output verbosity: 0=none, 1=low, 2=medium, 3=high
- **kwds :
Keywords arguments to be passed to the database instantiation method.
"""
Sampler.__init__(self, input, db, name, calc_deviance=calc_deviance, **kwds)
self._sm_assigned = False
self.step_method_dict = {}
for s in self.stochastics:
self.step_method_dict[s] = []
self._state = ['status', '_current_iter', '_iter', '_tune_interval', '_burn',
'_thin', '_tuned_count', '_tune_throughout', '_burn_till_tuned']
def use_step_method(self, step_method_class, *args, **kwds):
"""
M.use_step_method(step_method_class, *args, **kwds)
Example of usage: To handle stochastic A with a Metropolis instance,
M.use_step_method(Metropolis, A, sig=.1)
To subsequently get a reference to the new step method,
S = M.step_method_dict[A][0]
"""
new_method = step_method_class(*args, **kwds)
if self.verbose > 1:
print_('Using step method %s. Stochastics: ' % step_method_class.__name__)
for s in new_method.stochastics:
self.step_method_dict[s].append(new_method)
if self.verbose > 1:
print_('\t'+s.__name__)
if self._sm_assigned:
self.step_methods.add(new_method)
setattr(new_method, '_model', self)
def remove_step_method(self, step_method):
"""
Removes a step method.
"""
try:
for s in step_method.stochastics:
self.step_method_dict[s].remove(step_method)
if hasattr(self, "step_methods"):
self.step_methods.discard(step_method)
self._sm_assigned = False
except AttributeError:
for sm in step_method:
self.remove_step_method(sm)
def assign_step_methods(self, verbose=-1, draw_from_prior_when_possible = True):
"""
Make sure every stochastic variable has a step method. If not,
assign a step method from the registry.
"""
if not self._sm_assigned:
if draw_from_prior_when_possible:
# Assign dataless stepper first
last_gen = set([])
for s in self.stochastics - self.observed_stochastics:
if s._random is not None:
if len(s.extended_children)==0:
last_gen.add(s)
dataless, dataless_gens = crawl_dataless(set(last_gen), [last_gen])
if len(dataless):
new_method = DrawFromPrior(dataless, dataless_gens[::-1], verbose=verbose)
setattr(new_method, '_model', self)
for d in dataless:
if not d.observed:
self.step_method_dict[d].append(new_method)
if self.verbose > 1:
print_('Assigning step method %s to stochastic %s' % (new_method.__class__.__name__, d.__name__))
for s in self.stochastics:
# If not handled by any step method, make it a new step method using the registry
if len(self.step_method_dict[s])==0:
new_method = assign_method(s, verbose=verbose)
setattr(new_method, '_model', self)
self.step_method_dict[s].append(new_method)
if self.verbose > 1:
print_('Assigning step method %s to stochastic %s' % (new_method.__class__.__name__, s.__name__))
self.step_methods = set()
for s in self.stochastics:
self.step_methods |= set(self.step_method_dict[s])
for sm in self.step_methods:
if sm.tally:
for name in sm._tuning_info:
self._funs_to_tally[sm._id+'_'+name] = lambda name=name, sm=sm: getattr(sm, name)
else:
# Change verbosity for step methods
for sm_key in self.step_method_dict:
for sm in self.step_method_dict[sm_key]:
sm.verbose = verbose
self.restore_sm_state()
self._sm_assigned = True
def sample(self, iter, burn=0, thin=1, tune_interval=1000, tune_throughout=True,
save_interval=None, burn_till_tuned=False, stop_tuning_after=5,
verbose=0, progress_bar=True):
"""
sample(iter, burn, thin, tune_interval, tune_throughout, save_interval, verbose, progress_bar)
Initialize traces, run sampling loop, clean up afterward. Calls _loop.
:Parameters:
- iter : int
Total number of iterations to do
- burn : int
Variables will not be tallied until this many iterations are complete, default 0
- thin : int
Variables will be tallied at intervals of this many iterations, default 1
- tune_interval : int
Step methods will be tuned at intervals of this many iterations, default 1000
- tune_throughout : boolean
If true, tuning will continue after the burnin period (True); otherwise tuning
will halt at the end of the burnin period.
- save_interval : int or None
If given, the model state will be saved at intervals of this many iterations
- verbose : boolean
- progress_bar : boolean
Display progress bar while sampling.
- burn_till_tuned: boolean
If True the Sampler would burn samples until all step methods are tuned.
A tuned step methods is one that was not tuned for the last `stop_tuning_after` tuning intervals.
The burn-in phase will have a minimum of 'burn' iterations but could be longer if
tuning is needed. After the phase is done the sampler will run for another
(iter - burn) iterations, and will tally the samples according to the 'thin' argument.
This means that the total number of iteration is update throughout the sampling
procedure.
If burn_till_tuned is True it also overrides the tune_thorughout argument, so no step method
will be tuned when sample are being tallied.
- stop_tuning_adfter: int
the number of untuned successive tuning interval needed to be reach in order for
the burn-in phase to be done (If burn_till_tuned is True).
"""
self.assign_step_methods(verbose=verbose)
if burn > iter:
raise ValueError('Burn interval cannot be larger than specified number of iterations.')
self._n_tally = int(iter) - int(burn)
if burn_till_tuned:
self._stop_tuning_after = stop_tuning_after
tune_throughout = False
if verbose > 0:
print "burn_til_tuned is True. tune_throughout is set to False"
burn = int(max(burn, stop_tuning_after * tune_interval))
iter = self._n_tally + burn
self._iter = int(iter)
self._burn = int(burn)
self._thin = int(thin)
self._tune_interval = int(tune_interval)
self._tune_throughout = tune_throughout
self._burn_till_tuned = burn_till_tuned
self._save_interval = save_interval
length = max(int(np.floor((1.0*iter-burn)/thin)), 1)
self.max_trace_length = length
# Flags for tuning
self._tuning = True
self._tuned_count = 0
# Progress bar
self.pbar = None
if not verbose and progress_bar:
self.pbar = ProgressBar(self._iter)
# Run sampler
Sampler.sample(self, iter, length, verbose)
def _loop(self):
# Set status flag
self.status='running'
try:
while self._current_iter < self._iter and not self.status == 'halt':
if self.status == 'paused':
break
i = self._current_iter
# Update progress bar
if not (i+1) % 100 and self.pbar:
self.pbar.animate(i+1)
# Tune at interval
if i and not (i % self._tune_interval) and self._tuning:
self.tune()
#update _burn and _iter if needed
if self._burn_till_tuned and (self._tuned_count == 0):
new_burn = self._current_iter + int(self._stop_tuning_after * self._tune_interval)
self._burn = max(new_burn, self._burn);
self._iter = self._burn + self._n_tally
self.pbar = ProgressBar(self._iter)
self.pbar.animate(i+1)
# Manage burn-in
if i == self._burn:
if self.verbose>0:
print_('\nBurn-in interval complete')
if not self._tune_throughout:
self._tuning = False
# Tell all the step methods to take a step
for step_method in self.step_methods:
if self.verbose > 2:
print_('Step method %s stepping' % step_method._id)
# Step the step method
step_method.step()
# Record sample to trace, if appropriate
if i % self._thin == 0 and i >= self._burn:
self.tally()
if self._save_interval is not None:
if i % self._save_interval==0:
self.save_state()
# Periodically commit samples to backend
if not i % 1000:
self.commit()
# Increment interation
self._current_iter += 1
except KeyboardInterrupt:
self.status='halt'
finally:
# Stop progress bar
if self.pbar:
self.pbar.animate(self._iter)
if self.status == 'halt':
self._halt()
def tune(self):
"""
Tell all step methods to tune themselves.
"""
# =======================================
# = This is what makes price.py puke... =
# =======================================
# Only tune during burn-in
# if self._current_iter > self._burn:
# self._tuning = False
# return
if self.verbose > 0:
print_('\tTuning at iteration', self._current_iter)
# Initialize counter for number of tuning stochastics
tuning_count = 0
for step_method in self.step_methods:
verbose = self.verbose
if step_method.verbose > -1:
verbose = step_method.verbose
# Tune step methods
tuning_count += step_method.tune(verbose=self.verbose)
if verbose > 1:
print_('\t\tTuning step method %s, returned %i\n' %(step_method._id, tuning_count))
sys.stdout.flush()
if self._burn_till_tuned:
if not tuning_count:
# If no step methods needed tuning, increment count
self._tuned_count += 1
else:
# Otherwise re-initialize count
self._tuned_count = 0
# n consecutive clean intervals removed tuning
# n is equal to self._stop_tuning_after
if self._tuned_count == self._stop_tuning_after:
if self.verbose > 0: print_('\nFinished tuning')
self._tuning = False
def get_state(self):
"""
Return the sampler and step methods current state in order to
restart sampling at a later time.
"""
self.step_methods = set()
for s in self.stochastics:
self.step_methods |= set(self.step_method_dict[s])
state = Sampler.get_state(self)
state['step_methods'] = {}
# The state of each StepMethod.
for sm in self.step_methods:
state['step_methods'][sm._id] = sm.current_state().copy()
return state
def restore_sm_state(self):
sm_state = self.db.getstate()
if sm_state is not None:
sm_state = sm_state.get('step_methods', {})
# Restore stepping methods state
for sm in self.step_methods:
sm.__dict__.update(sm_state.get(sm._id, {}))
def _calc_dic(self):
"""Calculates deviance information Criterion"""
# Find mean deviance
mean_deviance = np.mean(self.db.trace('deviance')(), axis=0)
# Set values of all parameters to their mean
for stochastic in self.stochastics:
# Calculate mean of paramter
try:
mean_value = np.mean(self.db.trace(stochastic.__name__)(), axis=0)
# Set current value to mean
stochastic.value = mean_value
except KeyError:
print_("No trace available for %s. DIC value may not be valid." % stochastic.__name__)
# Return twice deviance minus deviance at means
return 2*mean_deviance - self.deviance
# Make dic a property
def _get_dic(self):
return self._calc_dic()
dic = property(_get_dic)
|