def _log_likelihood(self,
phi,
alleleConfig,
baseline,
maxCopyNumber,
update_tree=True):
if update_tree:
##################################################
# some useful info about the tree,
# used by CNV related computations,
set_node_height(self.tssb)
set_path_from_root_to_node(self.tssb)
map_datum_to_node(self.tssb)
##################################################
# 注意：此处应该不受CN\genotype的限制，但不记录目标的CN和genotype
# 因为此处的parameter不是准确的phi,所以无法达到最优，但为了能够抽样得到
# 最佳结构。此处要使CN和genotype自由发挥
# 在时间上的先后顺序能够明确影响测序数据的位置，只有重叠位置的时候才会
# 发生
# 注意此处可以设置默认参数
if self.fixedC < 0:
return self.__log_likelihood_RD_BAF(phi, alleleConfig, baseline,
maxCopyNumber)
elif self.fixedC >= 0:
return self.__log_likelihood_RD_BAF(phi, alleleConfig, baseline,
maxCopyNumber, self.fixedC)
else:
raise Exception("fixedC is abnormal")
def _log_likelihood_GivenC(self,
phi,
C,
alleleConfig,
baseline,
maxCopyNumber,
update_tree=False):
if update_tree:
##################################################
# some useful info about the tree,
# used by CNV related computations,
set_node_height(self.tssb)
set_path_from_root_to_node(self.tssb)
map_datum_to_node(self.tssb)
##################################################
copyNumbers = None
if self.tag == "BASELINE":
copyNumbers = [2]
elif get_loga(self) > baseline:
copyNumbers = range(3, maxCopyNumber + 1)
else:
copyNumbers = range(0, 2)
pass
if C not in copyNumbers:
return -float("Inf")
ll, pi = self._getLLStripe(C, phi, baseline, alleleConfig)
self.copyNumber = C
self.genotype = pi
self.phi = phi
return ll
pass
def do_mcmc(stateManager, backupManager, safeToExit, runSucceeded, config,
state, treeWriter, stripes, stripeNum, tmpDir):
startIter = state['last_iteration'] + 1
unwrittenTreeL = []
mcmcSampleTimesL = []
lastMcmcSampleTime = time.time()
# If --tmp-dir is not specified on the command line, it will by default be
# None, which will cause mkdtemp() to place this directory under the system's
# temporary directory. This is the desired behaviour.
config['tmp_dir'] = tempfile.mkdtemp(prefix='pwgsdataexchange.',
dir=tmpDir)
for iteration in range(startIter, state['sample_number']):
safeToExit.set()
if iteration < 0:
logmsg(iteration)
# Referring to tssb as local variable instead of dictionary element is much
# faster.
tssb = state['tssb']
tssb.resample_assignments()
tssb.cull_tree()
# assign node ids
wts, nodes = tssb.get_mixture()
for i, node in enumerate(nodes):
node.id = i
##################################################
# some useful info about the tree,
# used by CNV related computations,
# to be called only after resampling assignments
set_node_height(tssb)
set_path_from_root_to_node(tssb)
map_datum_to_node(tssb)
##################################################
state['mh_acc'] = metropolis(tssb, state['mh_itr'], state['mh_std'],
state['mh_burnin'], stripeNum,
state['stripes_file'], state['rand_seed'],
config['tmp_dir'])
if float(state['mh_acc']) < 0.08 and state['mh_std'] < 10000:
state['mh_std'] = state['mh_std'] * 2.0
logmsg("Shrinking MH proposals. Now %f" % state['mh_std'])
if float(state['mh_acc']) > 0.5 and float(state['mh_acc']) < 0.99:
state['mh_std'] = state['mh_std'] / 2.0
logmsg("Growing MH proposals. Now %f" % state['mh_std'])
tssb.resample_sticks()
tssb.resample_stick_orders()
tssb.resample_hypers(dpAlpha=True, alphaDecay=True, dpGamma=True)
lastLlh = tssb.complete_data_log_likelihood()
if iteration >= 0:
state['cd_llh_traces'][iteration] = lastLlh
if True or mod(iteration, 10) == 0:
weights, nodes = tssb.get_mixture()
logmsg(' '.join([
str(v) for v in (iteration, len(nodes),
state['cd_llh_traces'][iteration],
state['mh_acc'], tssb.dpAlpha,
tssb.dpGamma, tssb.alphaDecay)
]))
if argmax(state['cd_llh_traces'][:iteration + 1]) == iteration:
logmsg("%f is best per-data complete data likelihood so far." %
(state['cd_llh_traces'][iteration]))
else:
state['burnin_cd_llh_traces'][
iteration + state['burnin_sample_number']] = lastLlh
# Can't just put tssb in unwrittenTreeL, as this object will be modified
# on subsequent iterations, meaning any stored references in
# unwrittenTreeL will all point to the same sample.
serialized = pickle.dumps(tssb, protocol=pickle.HIGHEST_PROTOCOL)
unwrittenTreeL.append((serialized, iteration, lastLlh))
state['tssb'] = tssb
state['rand_state'] = get_state()
state['last_iteration'] = iteration
if len([
C for C in state['tssb'].root['children']
if C['node'].has_data()
]) > 1:
logmsg('Polyclonal tree detected with %s clones.' %
len(state['tssb'].root['children']))
newMcmcSampleTime = time.time()
mcmcSampleTimesL.append(newMcmcSampleTime - lastMcmcSampleTime)
lastMcmcSampleTime = newMcmcSampleTime
# It's not safe to exit while performing file IO, as we don't want
# trees.zip or the computation state file to become corrupted from an
# interrupted write.
safeToExit.clear()
shouldWriteBackup = iteration % state[
'write_backups_every'] == 0 and iteration != startIter
shouldWriteState = iteration % state['write_state_every'] == 0
isLastIteration = (iteration == state['sample_number'] - 1)
# If backup is scheduled to be written, write both it and full program
# state regardless of whether we're scheduled to write state this
# iteration.
if shouldWriteBackup or shouldWriteState or isLastIteration:
with open('mcmc_samples.txt', 'a') as mcmcf:
llhsAndTimes = [(itr, llh, itr_time) for (
tssb, itr,
llh), itr_time in zip(unwrittenTreeL, mcmcSampleTimesL)]
llhsAndTimes = '\n'.join([
'%s\t%s\t%s' % (itr, llh, itr_time)
for itr, llh, itr_time in llhsAndTimes
])
mcmcf.write(llhsAndTimes + '\n')
treeWriter.write_trees(unwrittenTreeL)
stateManager.write_state(state)
unwrittenTreeL = []
mcmcSampleTimesL = []
if shouldWriteBackup:
backupManager.save_backup()
backupManager.remove_backup()
safeToExit.clear()
# save the best tree
print_top_trees(TreeWriter.defaultArchiveFn, state['top_k_trees_file'],
state['top_k'])
# save clonal frequencies
freq = dict([(g, []) for g in state['stripe_list']])
stripeL = array(freq.keys(), str)
stripeL.shape = (1, len(stripeL))
savetxt(state['clonal_freqs_file'],
vstack((stripeL, array([freq[g] for g in freq.keys()]).T)),
fmt='%s',
delimiter=', ')
safeToExit.set()
runSucceeded.set()