def test_mk_new_dir(self):
self.assertFalse(os.path.exists(self.path))
functions.mk_new_dir(self.path)
self.assertTrue(os.path.exists(self.path))
self.assertTrue(os.path.isdir(self.path))
for i in range(10):
functions.mk_new_dir(self.path)
p = self.path + '-' + str(i)
self.assertTrue(os.path.exists(p))
self.assertTrue(os.path.isdir(p))
self.register_dir(p)
def test_mk_new_dir(self):
self.assertFalse(os.path.exists(self.path))
functions.mk_new_dir(self.path)
self.assertTrue(os.path.exists(self.path))
self.assertTrue(os.path.isdir(self.path))
for i in range(10):
functions.mk_new_dir(self.path)
p = self.path + '-' + str(i)
self.assertTrue(os.path.exists(p))
self.assertTrue(os.path.isdir(p))
self.register_dir(p)
def main_cli(argv = sys.argv):
description = '{name} {version}'.format(**_program_info)
parser = argparse.ArgumentParser(description = description,
formatter_class = argparse_utils.SmartHelpFormatter)
parser.add_argument('-o', '--observed-configs',
nargs = '+',
type = argparse_utils.arg_is_config,
required = True,
help = ('One or more msBayes config files to be used to either '
'calculate or simulate observed summary statistics. If '
'used in combination with `-r` each config will be used to '
'simulate pseudo-observed data. If analyzing real data, do '
'not use the `-r` option, and the fasta files specified '
'within the config must exist and contain the sequence '
'data.'))
parser.add_argument('-p', '--prior-configs',
nargs = '+',
type = argparse_utils.arg_is_path,
required = True,
help = ('One or more config files to be used to generate prior '
'samples. If more than one config is specified, they '
'should be separated by spaces. '
'This option can also be used to specify the path to a '
'directory containing the prior samples and summary '
'statistic means and standard deviations generated by a '
'previous run using the `generate-samples-only` option. '
'These files should be found in the directory '
'`pymsbayes-output/prior-stats-summaries`. The'
'`pymsbayes-output/model-key.txt` also needs to be present.'
' If specifying this directory, it should be the only '
'argument (i.e., no other directories or config files can '
'be provided).'))
parser.add_argument('-r', '--reps',
action = 'store',
type = argparse_utils.arg_is_nonnegative_int,
default = 0,
help = ('This option has two effects. First, it signifies that '
'the analysis will be simulation based (i.e., no real '
'data will be used). Second, it specifies how many '
'simulation replicates to perform (i.e., how many data '
'sets to simulate and analyze).'))
parser.add_argument('-n', '--num-prior-samples',
action = 'store',
type = argparse_utils.arg_is_positive_int,
default = 1000000,
help = ('The number of prior samples to simulate for each prior '
'config specified with `-p`.'))
parser.add_argument('--prior-batch-size',
action = 'store',
type = argparse_utils.arg_is_positive_int,
default = 10000,
help = ('The number of prior samples to simulate for each batch.'))
parser.add_argument('--generate-samples-only',
action = 'store_true',
help = ('Only generate samples from models as requested. I.e., '
'No analyses are performed to approximate posteriors. '
'This option can be useful if you want the prior samples '
'for other purposes.'))
parser.add_argument('--num-posterior-samples',
action = 'store',
type = argparse_utils.arg_is_positive_int,
default = 1000,
help = ('The number of posterior samples desired for each '
'analysis. Default: 1000.'))
parser.add_argument('--num-standardizing-samples',
action = 'store',
type = argparse_utils.arg_is_positive_int,
default = 10000,
help = ('The number of prior samples desired to use for '
'standardizing statistics. Default: 10000.'))
parser.add_argument('--np',
action = 'store',
type = argparse_utils.arg_is_positive_int,
default = multiprocessing.cpu_count(),
help = ('The maximum number of processes to run in parallel. The '
'default is the number of CPUs available on the machine.'))
parser.add_argument('--output-dir',
action = 'store',
type = argparse_utils.arg_is_dir,
help = ('The directory in which all output files will be written. '
'The default is to use the directory of the first observed '
'config file.'))
parser.add_argument('--temp-dir',
action = 'store',
type = argparse_utils.arg_is_dir,
help = ('A directory to temporarily stage files. The default is to '
'use the output directory.'))
parser.add_argument('--staging-dir',
action = 'store',
type = argparse_utils.arg_is_dir,
help = ('A directory to temporarily stage prior files. This option '
'can be useful on clusters to speed up I/O while '
'generating prior samples. You can designate a local temp '
'directory on a compute node to avoid constant writing to '
'a shared drive. The default is to use the `temp-dir`.'))
parser.add_argument('-s', '--stat-prefixes',
nargs = '*',
type = str,
help = ('Prefixes of summary statistics to use in the analyses. '
'The prefixes should be separated by spaces. '
'Default: `-s pi wattTheta pi.net tajD.denom`.'))
parser.add_argument('-b', '--bandwidth',
action = 'store',
type = float,
help = ('Smoothing parameter for the posterior kernal density '
'estimation. This option is used for the `glm` '
'regression method. The default is 2 / '
'`num-posterior-samples`.'))
parser.add_argument('-q', '--num-posterior-quantiles',
action = 'store',
type = argparse_utils.arg_is_positive_int,
default = 1000,
help = ('The number of equally spaced quantiles at which to '
'evaluate the GLM-estimated posterior density. '
'Default: 1000.'))
parser.add_argument('--reporting-frequency',
action = 'store',
type = argparse_utils.arg_is_nonnegative_int,
default = 0,
help = ('Suggested frequency (in number of prior samples) for '
'running regression and reporting current results. '
'Default: 0 (only report final results). '
'If a value is given, it may be adjusted so that the '
'reporting frequency is a multiple of the multi-processed '
'batch size.'))
parser.add_argument('--sort-index',
action = 'store',
type = argparse_utils.arg_is_nonnegative_int,
default = 0,
choices = range(12),
help = argparse_utils.get_sort_index_help_message())
parser.add_argument('--no-global-estimate',
action = 'store_true',
help = ('If multiple prior models are specified, by default a '
'global estimate is performed averaging over all models. '
'This option prevents the global estimation (i.e., only '
'inferences for each model are made).'))
parser.add_argument('--compress',
action = 'store_true',
help = 'Compress large results files.')
parser.add_argument('--keep-temps',
action = 'store_true',
help = 'Keep all temporary files.')
parser.add_argument('--seed',
action = 'store',
type = int,
help = 'Random number seed to use for the analysis.')
parser.add_argument('--output-prefix',
action = 'store',
type = str,
default = '',
help = ('Prefix to use at beginning of output files. The default '
'is no prefix.'))
parser.add_argument('--data-key-path',
action = 'store',
type = argparse_utils.arg_is_file,
help = ('The path to a `data-key.txt` file generated by a previous '
'run. This file should be found in the directory '
'`pymsbayes-output/data-key.txt`. This option '
'will override the `-o`/`--observed-configs` option, and '
'is intended to be used in combination with the '
'`--start-from` option to restart an analysis.'))
parser.add_argument('--start-from-simulation-index',
action = 'store',
type = argparse_utils.arg_is_nonnegative_int,
default = 0,
help = ('The simulation index at which to begin analyses. Must be '
'used in combination with either the number of simulation '
'replicates (`-r`/`--reps`) or the `--data-key-path` '
'option, and must be a positive '
'integer that is less than the number of simulation '
'replicates. This option can be useful if an analysis '
'needs to be restarted.'))
parser.add_argument('--start-from-observed-index',
action = 'store',
type = argparse_utils.arg_is_nonnegative_int,
default = 0,
help = ('The observed config index at which to begin analyses. '
'Can be used in combination with the `--data-key-path` '
'option to restart long-running, multi-observed-config '
'analyses'))
parser.add_argument('--dry-run',
action = 'store_true',
help = 'Do not run analyses; only process settings')
parser.add_argument('--version',
action = 'version',
version = '%(prog)s ' + _program_info['version'],
help = 'Report version and exit.')
parser.add_argument('--quiet',
action = 'store_true',
help = 'Run without verbose messaging.')
parser.add_argument('--debug',
action = 'store_true',
help = 'Run in debugging mode.')
if argv == sys.argv:
args = parser.parse_args()
else:
args = parser.parse_args(argv)
##########################################################################
## handle args
from pymsbayes.utils.messaging import (LoggingControl,
InfoLogger)
LoggingControl.set_logging_level("INFO")
if args.quiet:
LoggingControl.set_logging_level("WARNING")
if args.debug:
LoggingControl.set_logging_level("DEBUG")
log = LoggingControl.get_logger(__name__)
from pymsbayes.workers import (MsBayesWorker, merge_prior_files,
ObsSumStatsWorker)
from pymsbayes.teams import ABCTeam
from pymsbayes.utils.functions import (is_file, is_dir, long_division,
mk_new_dir)
from pymsbayes.utils.parsing import (get_patterns_from_prefixes,
DEFAULT_STAT_PATTERNS, DIV_MODEL_PATTERNS, MODEL_PATTERNS,
PSI_PATTERNS, MEAN_TAU_PATTERNS, OMEGA_PATTERNS, CV_PATTERNS,
line_count)
from pymsbayes.utils import sumresults, errors
from pymsbayes.manager import Manager
from pymsbayes.utils.tempfs import TempFileSystem
from pymsbayes.config import MsBayesConfig
from pymsbayes.utils import (GLOBAL_RNG, set_memory_trace,
MSBAYES_SORT_INDEX, ToolPathManager)
MSBAYES_SORT_INDEX.set_index(args.sort_index)
if len(args.observed_configs) != len(set(args.observed_configs)):
raise ValueError('All paths to observed config files must be unique')
if args.num_standardizing_samples > args.num_prior_samples:
args.num_standardizing_samples = args.num_prior_samples
# get full paths to tools
msbayes_path = ToolPathManager.get_tool_full_path('msbayes.pl')
dpp_msbayes_path = ToolPathManager.get_tool_full_path('dpp-msbayes.pl')
eureject_path = ToolPathManager.get_tool_full_path('eureject')
abctb_path = ToolPathManager.get_tool_full_path('ABCestimator')
# vet prior-configs option
using_previous_priors = False
previous_prior_dir = None
if (len(args.prior_configs) == 1) and (is_dir(args.prior_configs[0])):
previous_prior_dir = args.prior_configs.pop(0)
previous_priors = glob.glob(os.path.join(previous_prior_dir,
'*-prior-sample.txt'))
previous_sums = glob.glob(os.path.join(previous_prior_dir,
'*-means-and-std-devs.txt'))
if (not previous_priors) or (not previous_sums):
raise ValueError('directory {0!r} specified with `prior-configs` '
'option does not contain necessary prior and summary '
'files'.format(args.prior_configs[0]))
using_previous_priors = True
else:
for path in args.prior_configs:
if not is_file(path):
raise ValueError('prior config {0!r} is not a file'.format(
path))
if len(args.prior_configs) != len(set(args.prior_configs)):
raise ValueError('All paths to prior config files must be unique')
if not args.output_dir:
args.output_dir = os.path.dirname(args.observed_configs[0])
base_dir = mk_new_dir(os.path.join(args.output_dir, 'pymsbayes-results'))
if not args.temp_dir:
args.temp_dir = base_dir
info_path = os.path.join(base_dir, args.output_prefix + \
'pymsbayes-info.txt')
info = InfoLogger(info_path)
info.write('[pymsbayes]'.format(base_dir))
info.write('\tversion = {version}'.format(**_program_info))
info.write('\toutput_directory = {0}'.format(base_dir))
temp_fs = TempFileSystem(parent=args.temp_dir, prefix='temp-files-')
base_temp_dir = temp_fs.base_dir
info.write('\ttemp_directory = {0}'.format(base_temp_dir))
info.write('\tsort_index = {0}'.format(
MSBAYES_SORT_INDEX.current_value()))
info.write('\tsimulation_reps = {0}'.format(args.reps))
stat_patterns = DEFAULT_STAT_PATTERNS
if args.stat_prefixes:
for i in range(len(args.stat_prefixes)):
if not args.stat_prefixes[i].endswith('.'):
args.stat_prefixes[i] += '.'
stat_patterns = get_patterns_from_prefixes(
args.stat_prefixes,
ignore_case=True)
if not args.bandwidth:
args.bandwidth = 2 / float(args.num_posterior_samples)
if not args.seed:
args.seed = random.randint(1, 999999999)
GLOBAL_RNG.seed(args.seed)
if args.data_key_path:
observed_map = sumresults.parse_data_key_file(args.data_key_path)
observed_paths = [observed_map[k] for k in sorted(observed_map.keys())]
else:
observed_dir = mk_new_dir(os.path.join(base_dir,
'observed-summary-stats'))
observed_paths = [os.path.join(observed_dir, args.output_prefix + \
'observed-{0}.txt'.format(i+1)) for i in range(len(
args.observed_configs))]
info.write('\tseed = {0}'.format(args.seed))
info.write('\tnum_processors = {0}'.format(args.np))
info.write('\tnum_prior_samples = {0}'.format(
args.num_prior_samples))
info.write('\tnum_standardizing_samples = {0}'.format(
args.num_standardizing_samples))
info.write('\tbandwidth = {0}'.format(args.bandwidth))
info.write('\tposterior_quantiles = {0}'.format(
args.num_posterior_quantiles))
info.write('\tposterior_sample_size = {0}'.format(
args.num_posterior_samples))
info.write('\tstat_patterns = {0}'.format(
', '.join([p.pattern for p in stat_patterns])))
# vet observed configs
ref_config_path = args.observed_configs[0]
ref_config = MsBayesConfig(ref_config_path)
all_config_paths = []
num_taxon_pairs = ref_config.npairs
assert num_taxon_pairs > 0
for config in args.observed_configs:
all_config_paths.append(config)
if not ref_config.equal_sample_table(config):
if not args.keep_temps:
temp_fs.purge()
raise errors.SampleTableError(
'sample tables in config {0!r} and {1!r} differ; '
'all sample tables must be the same.'.format(
ref_config_path, config))
info.write('\tnum_taxon_pairs = {0}'.format(num_taxon_pairs))
info.write('\tdry_run = {0}'.format(args.dry_run))
info.write('\t[[tool_paths]]')
info.write('\t\tdpp_msbayes = {0}'.format(dpp_msbayes_path))
info.write('\t\tmsbayes = {0}'.format(msbayes_path))
info.write('\t\teureject = {0}'.format(eureject_path))
info.write('\t\tabcestimator = {0}'.format(abctb_path))
info.write('\t[[observed_configs]]')
for i, cfg in enumerate(args.observed_configs):
info.write('\t\t{0} = {1}'.format(i + 1, os.path.relpath(cfg,
os.path.dirname(info_path))))
abc_team = ABCTeam(
temp_fs = temp_fs,
observed_stats_files = observed_paths,
num_taxon_pairs = num_taxon_pairs,
config_paths = args.prior_configs,
previous_prior_dir = previous_prior_dir,
num_prior_samples = args.num_prior_samples,
num_processors = args.np,
num_standardizing_samples = args.num_standardizing_samples,
num_posterior_samples = args.num_posterior_samples,
num_posterior_density_quantiles = args.num_posterior_quantiles,
batch_size = args.prior_batch_size,
output_dir = base_dir,
output_prefix = args.output_prefix,
prior_temp_dir = args.staging_dir,
rng = GLOBAL_RNG,
report_parameters = True,
stat_patterns = stat_patterns,
eureject_exe_path = eureject_path,
abctoolbox_exe_path = abctb_path,
msbayes_exe_path = None,
abctoolbox_bandwidth = args.bandwidth,
omega_threshold = 0.01,
cv_threshold = 0.01,
compress = args.compress,
reporting_frequency = args.reporting_frequency,
keep_temps = args.keep_temps,
global_estimate_only = False,
global_estimate = not args.no_global_estimate,
generate_prior_samples_only = args.generate_samples_only,
start_from_simulation_index = args.start_from_simulation_index,
start_from_observed_index = args.start_from_observed_index)
models_to_configs = {}
configs_to_models = {}
for k, v in abc_team.models.iteritems():
models_to_configs[k] = v
configs_to_models[v] = k
cfg = MsBayesConfig(v)
all_config_paths.append(v)
# vet prior configs
if not ref_config.equal_sample_table(cfg):
if not args.keep_temps:
temp_fs.purge()
raise errors.SampleTableError(
'sample tables in config {0!r} and {1!r} differ; '
'all sample tables must be the same.'.format(
ref_config_path, v))
info.write('\t[[observed_paths]]')
for i in sorted(abc_team.observed_stats_paths.iterkeys()):
info.write('\t\t{0} = {1}'.format(i, os.path.relpath(
abc_team.observed_stats_paths[i],
os.path.dirname(info_path))))
info.write('\t[[prior_configs]]')
for i in sorted(abc_team.models.iterkeys()):
info.write('\t\t{0} = {1}'.format(i, os.path.relpath(
abc_team.models[i],
os.path.dirname(info_path))))
##########################################################################
## begin analysis --- get observed summary stats
set_memory_trace() # start logging memory profile
start_time = datetime.datetime.now()
if args.data_key_path:
log.info('Using provided summary statitics...')
elif not args.dry_run:
obs_temp_dir = base_temp_dir
if args.staging_dir:
obs_temp_dir = args.staging_dir
observed_temp_fs = TempFileSystem(parent = obs_temp_dir,
prefix = 'observed-temps-')
if args.reps < 1:
log.info('Calculating summary statistics from sequence data...')
obs_workers = []
for i, cfg in enumerate(args.observed_configs):
ss_worker = ObsSumStatsWorker(
temp_fs = observed_temp_fs,
config_path = cfg,
output_path = observed_paths[i],
schema = 'abctoolbox',
stat_patterns = stat_patterns)
obs_workers.append(ss_worker)
obs_workers = Manager.run_workers(
workers = obs_workers,
num_processors = args.np)
# re-vet all configs to see if some were changed by obsSumStats.pl
new_ref_config = ref_config
ref_modified = False
# new ref because if all configs were updated all is good
if not ref_config.equal_sample_table(ref_config_path):
ref_modified = True
new_ref_config = MsBayesConfig(ref_config_path)
log.warning("""
The alignment lengths in config
{0!r}
have been corrected for sites with *any* ambiguous bases and/or gaps by
obsSumStats.pl.
""".format(ref_config_path))
for config in all_config_paths:
if not new_ref_config.equal_sample_table(config):
corrected_config = config
if ref_modified:
corrected_config = ref_config_path
if not args.keep_temps:
observed_temp_fs.purge()
temp_fs.purge()
raise errors.SampleTableError("""
The sample tables in configs
{0!r}
and
{1!r}
differ because obsSumStats.pl modified alignment lengths in config
{2!r}
to correct for sites in the alignments with *any* ambiguous bases and/or gaps.
Please make sure the sample tables in all configs will be the same after
correcting alignment lengths for sites that contain *any* ambiguous bases
and/or gaps. You can do this by copying and pasting the sample table in
{2!r}
that has been corrected by obsSumStats.pl into the other configs that were not
corrected.
""".format(ref_config_path, config, corrected_config))
else:
log.info('Simulating summary statistics from observed configs...')
num_observed_workers = min([args.reps, args.np])
if args.reps <= args.np:
observed_batch_size = 1
remainder = 0
else:
observed_batch_size, remainder = long_division(args.reps,
args.np)
msbayes_workers = []
for idx, cfg in enumerate(args.observed_configs):
observed_model_idx = configs_to_models.get(cfg,
None)
schema = 'abctoolbox'
for i in range(num_observed_workers):
worker = MsBayesWorker(
temp_fs = observed_temp_fs,
sample_size = observed_batch_size,
config_path = cfg,
model_index = observed_model_idx,
report_parameters = True,
schema = schema,
include_header = True,
stat_patterns = stat_patterns,
write_stats_file = False,
staging_dir = None,
tag = idx)
msbayes_workers.append(worker)
if remainder > 0:
worker = MsBayesWorker(
temp_fs = observed_temp_fs,
sample_size = remainder,
config_path = cfg,
model_index = observed_model_idx,
report_parameters = True,
schema = schema,
include_header = True,
stat_patterns = stat_patterns,
write_stats_file = False,
staging_dir = None,
tag = idx)
msbayes_workers.append(worker)
# run parallel msbayes processes
msbayes_workers = Manager.run_workers(
workers = msbayes_workers,
num_processors = args.np)
workers = dict(zip(range(len(args.observed_configs)),
[[] for i in range(len(args.observed_configs))]))
for w in msbayes_workers:
workers[w.tag].append(w)
# merge simulated observed data into one file
for i in range(len(args.observed_configs)):
merge_prior_files([w.prior_path for w in workers[i]],
observed_paths[i])
lc = line_count(observed_paths[i], ignore_headers=True)
if lc != args.reps:
if not args.keep_temps:
temp_fs.purge()
raise Exception('The number of observed simulations ({0}) '
'generated for observed config {1!r} and output to '
'file {2!r} does not match the number of reps '
'({3})'.format(lc, args.observed_configs[i],
observed_paths[i], args.reps))
if not args.keep_temps:
log.debug('purging observed temps...')
observed_temp_fs.purge()
##########################################################################
## Begin ABC analyses
if not args.dry_run:
abc_team.run()
stop_time = datetime.datetime.now()
log.info('Done!')
info.write('\t[[run_stats]]', log.info)
info.write('\t\tstart_time = {0}'.format(str(start_time)), log.info)
info.write('\t\tstop_time = {0}'.format(str(stop_time)), log.info)
info.write('\t\ttotal_duration = {0}'.format(str(stop_time - start_time)),
log.info)
if not args.keep_temps:
log.debug('purging temps...')
temp_fs.purge()
def main_cli(argv=sys.argv):
description = '{name} {version}'.format(**_program_info)
parser = argparse.ArgumentParser(
description=description,
formatter_class=argparse_utils.SmartHelpFormatter)
parser.add_argument(
'-o',
'--observed-configs',
nargs='+',
type=argparse_utils.arg_is_config,
required=True,
help=('One or more msBayes config files to be used to either '
'calculate or simulate observed summary statistics. If '
'used in combination with `-r` each config will be used to '
'simulate pseudo-observed data. If analyzing real data, do '
'not use the `-r` option, and the fasta files specified '
'within the config must exist and contain the sequence '
'data.'))
parser.add_argument(
'-p',
'--prior-configs',
nargs='+',
type=argparse_utils.arg_is_path,
required=True,
help=('One or more config files to be used to generate prior '
'samples. If more than one config is specified, they '
'should be separated by spaces. '
'This option can also be used to specify the path to a '
'directory containing the prior samples and summary '
'statistic means and standard deviations generated by a '
'previous run using the `generate-samples-only` option. '
'These files should be found in the directory '
'`pymsbayes-output/prior-stats-summaries`. The'
'`pymsbayes-output/model-key.txt` also needs to be present.'
' If specifying this directory, it should be the only '
'argument (i.e., no other directories or config files can '
'be provided).'))
parser.add_argument(
'-r',
'--reps',
action='store',
type=argparse_utils.arg_is_nonnegative_int,
default=0,
help=('This option has two effects. First, it signifies that '
'the analysis will be simulation based (i.e., no real '
'data will be used). Second, it specifies how many '
'simulation replicates to perform (i.e., how many data '
'sets to simulate and analyze).'))
parser.add_argument(
'-n',
'--num-prior-samples',
action='store',
type=argparse_utils.arg_is_positive_int,
default=1000000,
help=('The number of prior samples to simulate for each prior '
'config specified with `-p`.'))
parser.add_argument(
'--prior-batch-size',
action='store',
type=argparse_utils.arg_is_positive_int,
default=10000,
help=('The number of prior samples to simulate for each batch.'))
parser.add_argument(
'--generate-samples-only',
action='store_true',
help=('Only generate samples from models as requested. I.e., '
'No analyses are performed to approximate posteriors. '
'This option can be useful if you want the prior samples '
'for other purposes.'))
parser.add_argument(
'--num-posterior-samples',
action='store',
type=argparse_utils.arg_is_positive_int,
default=1000,
help=('The number of posterior samples desired for each '
'analysis. Default: 1000.'))
parser.add_argument('--num-standardizing-samples',
action='store',
type=argparse_utils.arg_is_positive_int,
default=10000,
help=('The number of prior samples desired to use for '
'standardizing statistics. Default: 10000.'))
parser.add_argument(
'--np',
action='store',
type=argparse_utils.arg_is_positive_int,
default=multiprocessing.cpu_count(),
help=('The maximum number of processes to run in parallel. The '
'default is the number of CPUs available on the machine.'))
parser.add_argument(
'--output-dir',
action='store',
type=argparse_utils.arg_is_dir,
help=('The directory in which all output files will be written. '
'The default is to use the directory of the first observed '
'config file.'))
parser.add_argument(
'--temp-dir',
action='store',
type=argparse_utils.arg_is_dir,
help=('A directory to temporarily stage files. The default is to '
'use the output directory.'))
parser.add_argument(
'--staging-dir',
action='store',
type=argparse_utils.arg_is_dir,
help=('A directory to temporarily stage prior files. This option '
'can be useful on clusters to speed up I/O while '
'generating prior samples. You can designate a local temp '
'directory on a compute node to avoid constant writing to '
'a shared drive. The default is to use the `temp-dir`.'))
parser.add_argument(
'-s',
'--stat-prefixes',
nargs='*',
type=str,
help=('Prefixes of summary statistics to use in the analyses. '
'The prefixes should be separated by spaces. '
'Default: `-s pi wattTheta pi.net tajD.denom`.'))
parser.add_argument(
'-b',
'--bandwidth',
action='store',
type=float,
help=('Smoothing parameter for the posterior kernal density '
'estimation. This option is used for the `glm` '
'regression method. The default is 2 / '
'`num-posterior-samples`.'))
parser.add_argument(
'-q',
'--num-posterior-quantiles',
action='store',
type=argparse_utils.arg_is_positive_int,
default=1000,
help=('The number of equally spaced quantiles at which to '
'evaluate the GLM-estimated posterior density. '
'Default: 1000.'))
parser.add_argument(
'--reporting-frequency',
action='store',
type=argparse_utils.arg_is_nonnegative_int,
default=0,
help=('Suggested frequency (in number of prior samples) for '
'running regression and reporting current results. '
'Default: 0 (only report final results). '
'If a value is given, it may be adjusted so that the '
'reporting frequency is a multiple of the multi-processed '
'batch size.'))
parser.add_argument('--sort-index',
action='store',
type=argparse_utils.arg_is_nonnegative_int,
default=0,
choices=range(12),
help=argparse_utils.get_sort_index_help_message())
parser.add_argument(
'--no-global-estimate',
action='store_true',
help=('If multiple prior models are specified, by default a '
'global estimate is performed averaging over all models. '
'This option prevents the global estimation (i.e., only '
'inferences for each model are made).'))
parser.add_argument('--compress',
action='store_true',
help='Compress large results files.')
parser.add_argument('--keep-temps',
action='store_true',
help='Keep all temporary files.')
parser.add_argument('--seed',
action='store',
type=int,
help='Random number seed to use for the analysis.')
parser.add_argument(
'--output-prefix',
action='store',
type=str,
default='',
help=('Prefix to use at beginning of output files. The default '
'is no prefix.'))
parser.add_argument(
'--data-key-path',
action='store',
type=argparse_utils.arg_is_file,
help=('The path to a `data-key.txt` file generated by a previous '
'run. This file should be found in the directory '
'`pymsbayes-output/data-key.txt`. This option '
'will override the `-o`/`--observed-configs` option, and '
'is intended to be used in combination with the '
'`--start-from` option to restart an analysis.'))
parser.add_argument(
'--start-from-simulation-index',
action='store',
type=argparse_utils.arg_is_nonnegative_int,
default=0,
help=('The simulation index at which to begin analyses. Must be '
'used in combination with either the number of simulation '
'replicates (`-r`/`--reps`) or the `--data-key-path` '
'option, and must be a positive '
'integer that is less than the number of simulation '
'replicates. This option can be useful if an analysis '
'needs to be restarted.'))
parser.add_argument(
'--start-from-observed-index',
action='store',
type=argparse_utils.arg_is_nonnegative_int,
default=0,
help=('The observed config index at which to begin analyses. '
'Can be used in combination with the `--data-key-path` '
'option to restart long-running, multi-observed-config '
'analyses'))
parser.add_argument('--dry-run',
action='store_true',
help='Do not run analyses; only process settings')
parser.add_argument('--version',
action='version',
version='%(prog)s ' + _program_info['version'],
help='Report version and exit.')
parser.add_argument('--quiet',
action='store_true',
help='Run without verbose messaging.')
parser.add_argument('--debug',
action='store_true',
help='Run in debugging mode.')
if argv == sys.argv:
args = parser.parse_args()
else:
args = parser.parse_args(argv)
##########################################################################
## handle args
from pymsbayes.utils.messaging import (LoggingControl, InfoLogger)
LoggingControl.set_logging_level("INFO")
if args.quiet:
LoggingControl.set_logging_level("WARNING")
if args.debug:
LoggingControl.set_logging_level("DEBUG")
log = LoggingControl.get_logger(__name__)
from pymsbayes.workers import (MsBayesWorker, merge_prior_files,
ObsSumStatsWorker)
from pymsbayes.teams import ABCTeam
from pymsbayes.utils.functions import (is_file, is_dir, long_division,
mk_new_dir)
from pymsbayes.utils.parsing import (get_patterns_from_prefixes,
DEFAULT_STAT_PATTERNS,
DIV_MODEL_PATTERNS, MODEL_PATTERNS,
PSI_PATTERNS, MEAN_TAU_PATTERNS,
OMEGA_PATTERNS, CV_PATTERNS,
line_count)
from pymsbayes.utils import sumresults, errors
from pymsbayes.manager import Manager
from pymsbayes.utils.tempfs import TempFileSystem
from pymsbayes.config import MsBayesConfig
from pymsbayes.utils import (GLOBAL_RNG, set_memory_trace,
MSBAYES_SORT_INDEX, ToolPathManager)
MSBAYES_SORT_INDEX.set_index(args.sort_index)
if len(args.observed_configs) != len(set(args.observed_configs)):
raise ValueError('All paths to observed config files must be unique')
if args.num_standardizing_samples > args.num_prior_samples:
args.num_standardizing_samples = args.num_prior_samples
# get full paths to tools
msbayes_path = ToolPathManager.get_tool_full_path('msbayes.pl')
dpp_msbayes_path = ToolPathManager.get_tool_full_path('dpp-msbayes.pl')
eureject_path = ToolPathManager.get_tool_full_path('eureject')
abctb_path = ToolPathManager.get_tool_full_path('ABCestimator')
# vet prior-configs option
using_previous_priors = False
previous_prior_dir = None
if (len(args.prior_configs) == 1) and (is_dir(args.prior_configs[0])):
previous_prior_dir = args.prior_configs.pop(0)
previous_priors = glob.glob(
os.path.join(previous_prior_dir, '*-prior-sample.txt'))
previous_sums = glob.glob(
os.path.join(previous_prior_dir, '*-means-and-std-devs.txt'))
if (not previous_priors) or (not previous_sums):
raise ValueError(
'directory {0!r} specified with `prior-configs` '
'option does not contain necessary prior and summary '
'files'.format(args.prior_configs[0]))
using_previous_priors = True
else:
for path in args.prior_configs:
if not is_file(path):
raise ValueError(
'prior config {0!r} is not a file'.format(path))
if len(args.prior_configs) != len(set(args.prior_configs)):
raise ValueError('All paths to prior config files must be unique')
if not args.output_dir:
args.output_dir = os.path.dirname(args.observed_configs[0])
base_dir = mk_new_dir(os.path.join(args.output_dir, 'pymsbayes-results'))
if not args.temp_dir:
args.temp_dir = base_dir
info_path = os.path.join(base_dir, args.output_prefix + \
'pymsbayes-info.txt')
info = InfoLogger(info_path)
info.write('[pymsbayes]'.format(base_dir))
info.write('\tversion = {version}'.format(**_program_info))
info.write('\toutput_directory = {0}'.format(base_dir))
temp_fs = TempFileSystem(parent=args.temp_dir, prefix='temp-files-')
base_temp_dir = temp_fs.base_dir
info.write('\ttemp_directory = {0}'.format(base_temp_dir))
info.write('\tsort_index = {0}'.format(MSBAYES_SORT_INDEX.current_value()))
info.write('\tsimulation_reps = {0}'.format(args.reps))
stat_patterns = DEFAULT_STAT_PATTERNS
if args.stat_prefixes:
for i in range(len(args.stat_prefixes)):
if not args.stat_prefixes[i].endswith('.'):
args.stat_prefixes[i] += '.'
stat_patterns = get_patterns_from_prefixes(args.stat_prefixes,
ignore_case=True)
if not args.bandwidth:
args.bandwidth = 2 / float(args.num_posterior_samples)
if not args.seed:
args.seed = random.randint(1, 999999999)
GLOBAL_RNG.seed(args.seed)
if args.data_key_path:
observed_map = sumresults.parse_data_key_file(args.data_key_path)
observed_paths = [observed_map[k] for k in sorted(observed_map.keys())]
else:
observed_dir = mk_new_dir(
os.path.join(base_dir, 'observed-summary-stats'))
observed_paths = [os.path.join(observed_dir, args.output_prefix + \
'observed-{0}.txt'.format(i+1)) for i in range(len(
args.observed_configs))]
info.write('\tseed = {0}'.format(args.seed))
info.write('\tnum_processors = {0}'.format(args.np))
info.write('\tnum_prior_samples = {0}'.format(args.num_prior_samples))
info.write('\tnum_standardizing_samples = {0}'.format(
args.num_standardizing_samples))
info.write('\tbandwidth = {0}'.format(args.bandwidth))
info.write('\tposterior_quantiles = {0}'.format(
args.num_posterior_quantiles))
info.write('\tposterior_sample_size = {0}'.format(
args.num_posterior_samples))
info.write('\tstat_patterns = {0}'.format(', '.join(
[p.pattern for p in stat_patterns])))
# vet observed configs
ref_config_path = args.observed_configs[0]
ref_config = MsBayesConfig(ref_config_path)
all_config_paths = []
num_taxon_pairs = ref_config.npairs
assert num_taxon_pairs > 0
for config in args.observed_configs:
all_config_paths.append(config)
if not ref_config.equal_sample_table(config):
if not args.keep_temps:
temp_fs.purge()
raise errors.SampleTableError(
'sample tables in config {0!r} and {1!r} differ; '
'all sample tables must be the same.'.format(
ref_config_path, config))
info.write('\tnum_taxon_pairs = {0}'.format(num_taxon_pairs))
info.write('\tdry_run = {0}'.format(args.dry_run))
info.write('\t[[tool_paths]]')
info.write('\t\tdpp_msbayes = {0}'.format(dpp_msbayes_path))
info.write('\t\tmsbayes = {0}'.format(msbayes_path))
info.write('\t\teureject = {0}'.format(eureject_path))
info.write('\t\tabcestimator = {0}'.format(abctb_path))
info.write('\t[[observed_configs]]')
for i, cfg in enumerate(args.observed_configs):
info.write('\t\t{0} = {1}'.format(
i + 1, os.path.relpath(cfg, os.path.dirname(info_path))))
abc_team = ABCTeam(
temp_fs=temp_fs,
observed_stats_files=observed_paths,
num_taxon_pairs=num_taxon_pairs,
config_paths=args.prior_configs,
previous_prior_dir=previous_prior_dir,
num_prior_samples=args.num_prior_samples,
num_processors=args.np,
num_standardizing_samples=args.num_standardizing_samples,
num_posterior_samples=args.num_posterior_samples,
num_posterior_density_quantiles=args.num_posterior_quantiles,
batch_size=args.prior_batch_size,
output_dir=base_dir,
output_prefix=args.output_prefix,
prior_temp_dir=args.staging_dir,
rng=GLOBAL_RNG,
report_parameters=True,
stat_patterns=stat_patterns,
eureject_exe_path=eureject_path,
abctoolbox_exe_path=abctb_path,
msbayes_exe_path=None,
abctoolbox_bandwidth=args.bandwidth,
omega_threshold=0.01,
cv_threshold=0.01,
compress=args.compress,
reporting_frequency=args.reporting_frequency,
keep_temps=args.keep_temps,
global_estimate_only=False,
global_estimate=not args.no_global_estimate,
generate_prior_samples_only=args.generate_samples_only,
start_from_simulation_index=args.start_from_simulation_index,
start_from_observed_index=args.start_from_observed_index)
models_to_configs = {}
configs_to_models = {}
for k, v in abc_team.models.iteritems():
models_to_configs[k] = v
configs_to_models[v] = k
cfg = MsBayesConfig(v)
all_config_paths.append(v)
# vet prior configs
if not ref_config.equal_sample_table(cfg):
if not args.keep_temps:
temp_fs.purge()
raise errors.SampleTableError(
'sample tables in config {0!r} and {1!r} differ; '
'all sample tables must be the same.'.format(
ref_config_path, v))
info.write('\t[[observed_paths]]')
for i in sorted(abc_team.observed_stats_paths.iterkeys()):
info.write('\t\t{0} = {1}'.format(
i,
os.path.relpath(abc_team.observed_stats_paths[i],
os.path.dirname(info_path))))
info.write('\t[[prior_configs]]')
for i in sorted(abc_team.models.iterkeys()):
info.write('\t\t{0} = {1}'.format(
i, os.path.relpath(abc_team.models[i],
os.path.dirname(info_path))))
##########################################################################
## begin analysis --- get observed summary stats
set_memory_trace() # start logging memory profile
start_time = datetime.datetime.now()
if args.data_key_path:
log.info('Using provided summary statitics...')
elif not args.dry_run:
obs_temp_dir = base_temp_dir
if args.staging_dir:
obs_temp_dir = args.staging_dir
observed_temp_fs = TempFileSystem(parent=obs_temp_dir,
prefix='observed-temps-')
if args.reps < 1:
log.info('Calculating summary statistics from sequence data...')
obs_workers = []
for i, cfg in enumerate(args.observed_configs):
ss_worker = ObsSumStatsWorker(temp_fs=observed_temp_fs,
config_path=cfg,
output_path=observed_paths[i],
schema='abctoolbox',
stat_patterns=stat_patterns)
obs_workers.append(ss_worker)
obs_workers = Manager.run_workers(workers=obs_workers,
num_processors=args.np)
# re-vet all configs to see if some were changed by obsSumStats.pl
new_ref_config = ref_config
ref_modified = False
# new ref because if all configs were updated all is good
if not ref_config.equal_sample_table(ref_config_path):
ref_modified = True
new_ref_config = MsBayesConfig(ref_config_path)
log.warning("""
The alignment lengths in config
{0!r}
have been corrected for sites with *any* ambiguous bases and/or gaps by
obsSumStats.pl.
""".format(ref_config_path))
for config in all_config_paths:
if not new_ref_config.equal_sample_table(config):
corrected_config = config
if ref_modified:
corrected_config = ref_config_path
if not args.keep_temps:
observed_temp_fs.purge()
temp_fs.purge()
raise errors.SampleTableError("""
The sample tables in configs
{0!r}
and
{1!r}
differ because obsSumStats.pl modified alignment lengths in config
{2!r}
to correct for sites in the alignments with *any* ambiguous bases and/or gaps.
Please make sure the sample tables in all configs will be the same after
correcting alignment lengths for sites that contain *any* ambiguous bases
and/or gaps. You can do this by copying and pasting the sample table in
{2!r}
that has been corrected by obsSumStats.pl into the other configs that were not
corrected.
""".format(ref_config_path, config, corrected_config))
else:
log.info('Simulating summary statistics from observed configs...')
num_observed_workers = min([args.reps, args.np])
if args.reps <= args.np:
observed_batch_size = 1
remainder = 0
else:
observed_batch_size, remainder = long_division(
args.reps, args.np)
msbayes_workers = []
for idx, cfg in enumerate(args.observed_configs):
observed_model_idx = configs_to_models.get(cfg, None)
schema = 'abctoolbox'
for i in range(num_observed_workers):
worker = MsBayesWorker(temp_fs=observed_temp_fs,
sample_size=observed_batch_size,
config_path=cfg,
model_index=observed_model_idx,
report_parameters=True,
schema=schema,
include_header=True,
stat_patterns=stat_patterns,
write_stats_file=False,
staging_dir=None,
tag=idx)
msbayes_workers.append(worker)
if remainder > 0:
worker = MsBayesWorker(temp_fs=observed_temp_fs,
sample_size=remainder,
config_path=cfg,
model_index=observed_model_idx,
report_parameters=True,
schema=schema,
include_header=True,
stat_patterns=stat_patterns,
write_stats_file=False,
staging_dir=None,
tag=idx)
msbayes_workers.append(worker)
# run parallel msbayes processes
msbayes_workers = Manager.run_workers(workers=msbayes_workers,
num_processors=args.np)
workers = dict(
zip(range(len(args.observed_configs)),
[[] for i in range(len(args.observed_configs))]))
for w in msbayes_workers:
workers[w.tag].append(w)
# merge simulated observed data into one file
for i in range(len(args.observed_configs)):
merge_prior_files([w.prior_path for w in workers[i]],
observed_paths[i])
lc = line_count(observed_paths[i], ignore_headers=True)
if lc != args.reps:
if not args.keep_temps:
temp_fs.purge()
raise Exception(
'The number of observed simulations ({0}) '
'generated for observed config {1!r} and output to '
'file {2!r} does not match the number of reps '
'({3})'.format(lc, args.observed_configs[i],
observed_paths[i], args.reps))
if not args.keep_temps:
log.debug('purging observed temps...')
observed_temp_fs.purge()
##########################################################################
## Begin ABC analyses
if not args.dry_run:
abc_team.run()
stop_time = datetime.datetime.now()
log.info('Done!')
info.write('\t[[run_stats]]', log.info)
info.write('\t\tstart_time = {0}'.format(str(start_time)), log.info)
info.write('\t\tstop_time = {0}'.format(str(stop_time)), log.info)
info.write('\t\ttotal_duration = {0}'.format(str(stop_time - start_time)),
log.info)
if not args.keep_temps:
log.debug('purging temps...')
temp_fs.purge()
