def __init__(self, path=None, environment=None, revision_cache=None):
"""
See :class:`lab.experiment.Experiment` for an explanation of
the *path* and *environment* parameters.
*revision_cache* is the directory for caching Fast Downward
revisions. It defaults to ``<scriptdir>/data/revision-cache``.
This directory can become very large since each revision uses
about 30 MB.
>>> from lab.environments import BaselSlurmEnvironment
>>> env = BaselSlurmEnvironment(email="*****@*****.**")
>>> exp = FastDownwardExperiment(environment=env)
You can add parsers with :meth:`.add_parser()`. See
:ref:`parsing` for how to write custom parsers and
:ref:`downward-parsers` for the list of built-in parsers. Which
parsers you should use depends on the algorithms you're running.
For single-search experiments, we recommend adding the following
parsers in this order:
>>> exp.add_parser(exp.EXITCODE_PARSER)
>>> exp.add_parser(exp.TRANSLATOR_PARSER)
>>> exp.add_parser(exp.SINGLE_SEARCH_PARSER)
>>> exp.add_parser(exp.PLANNER_PARSER)
"""
Experiment.__init__(self, path=path, environment=environment)
self.revision_cache = revision_cache or os.path.join(
get_default_data_dir(), 'revision-cache')
self._suites = defaultdict(list)
# Use OrderedDict to ensure that names are unique and ordered.
self._algorithms = OrderedDict()
self.add_command('remove-output-sas', ['rm', '-f', 'output.sas'])
def __init__(self, path=None, environment=None, revision_cache=None):
"""
See :class:`lab.experiment.Experiment` for an explanation of
the *path* and *environment* parameters.
*revision_cache* is the directory for caching Fast Downward
revisions. It defaults to ``<scriptdir>/data/revision-cache``.
This directory can become very large since each revision uses
about 30 MB.
>>> from lab.environments import MaiaEnvironment
>>> env = MaiaEnvironment(priority=-2)
>>> exp = FastDownwardExperiment(environment=env)
"""
Experiment.__init__(self, path=path, environment=environment)
self.revision_cache = revision_cache or os.path.join(
get_default_data_dir(), 'revision-cache')
self._suites = defaultdict(list)
# Use OrderedDict to ensure that names are unique and ordered.
self._algorithms = OrderedDict()
def __init__(self, path, repo, environment=None, combinations=None,
compact=True, limits=None, cache_dir=None):
"""
The experiment will be built at *path*.
*repo* must be the path to a Fast Downward repository. Among other things
this repository is used to search for benchmark files.
*environment* must be an :ref:`Environment <environments>` instance.
By default the experiment is run locally.
If given, *combinations* must be a list of :ref:`Checkout <checkouts>`
tuples of the form (Translator, Preprocessor, Planner). If combinations
is None (default), perform an experiment with the working copy in *repo*.
The *compact* parameter is only relevant for the search
stage. If *compact* is ``False``, the preprocessed task and
the two PDDL files are **copied** into the respective run
directories for all configurations. This requires a lot of
space (tens of GB), so it is strongly recommended to use the
default (``compact=True``) which only references these
files. Use ``compact=False`` only if you really need a
portable experiment.
If *limits* is given, it must be a dictionary that maps a
subset of the keys below to seconds and MiB. It will be used
to overwrite the default limits::
default_limits = {
'translate_time': 7200,
'translate_memory': 8192,
'preprocess_time': 7200,
'preprocess_memory': 8192,
'search_time': 1800,
'search_memory': 2048,
}
*cache_dir* is used to cache Fast Downward clones and preprocessed
tasks. By default it points to ``~/lab``.
.. note::
The directory *cache_dir* can grow very large (tens of GB).
Example: ::
repo = '/path/to/downward-repo'
env = GkiGridEnvironment(queue='xeon_core.q', priority=-2)
combos = [(Translator(repo, rev=123),
Preprocessor(repo, rev='e2a018c865f7'),
Planner(repo, rev='tip')]
exp = DownwardExperiment('/tmp/path', repo, environment=env,
combinations=combos,
limits={'search_time': 30,
'search_memory': 1024})
"""
Experiment.__init__(self, path, environment=environment, cache_dir=cache_dir)
#if not repo or not os.path.isdir(repo):
# logging.critical('The path "%s" is not a local Fast Downward '
# 'repository.' % repo)
self.repo = repo
self.orig_path = self.path
self.search_exp_path = self.path
self.preprocess_exp_path = self.path + '-p'
self._path_to_python = None
Checkout.REV_CACHE_DIR = os.path.join(self.cache_dir, 'revision-cache')
self.preprocessed_tasks_dir = os.path.join(self.cache_dir, 'preprocessed-tasks')
tools.makedirs(self.preprocessed_tasks_dir)
self.combinations = (combinations or
[(Translator(repo), Preprocessor(repo), Planner(repo))])
self.compact = compact
self.suites = defaultdict(list)
self._algorithms = []
self._portfolios = []
limits = limits or {}
for key, value in limits.items():
if key not in LIMITS:
logging.critical('Unknown limit: %s' % key)
self.limits = LIMITS
self.limits.update(limits)
# Save if this is a compact experiment i.e. preprocessed tasks are referenced.
self.set_property('compact', compact)
# TODO: Integrate this into the API.
self.include_preprocess_results_in_search_runs = True
self.compilation_options = ['-j%d' % self._jobs]
self._search_parsers = []
self.add_search_parser(os.path.join(DOWNWARD_SCRIPTS_DIR, 'search_parser.py'))
# Remove the default experiment steps
self.steps = Sequence()
self.add_step(Step('build-preprocess-exp', self.build, stage='preprocess'))
self.add_step(Step('run-preprocess-exp', self.run, stage='preprocess'))
self.add_fetcher(src=self.preprocess_exp_path,
dest=self.preprocessed_tasks_dir,
name='fetch-preprocess-results',
copy_all=True,
write_combined_props=False)
self.add_step(Step('build-search-exp', self.build, stage='search'))
self.add_PAC_fetcher(src='/home/sternron/gal-dreiman/downward/lab/examples/PAC_Preprocess_Output-eval/preprocess',#TODO change to be parameter
dest=self.search_exp_path,
name='fetch-preprocess-results',
copy_all=True,
write_combined_props=False)#new featcher to copy preprocess for PAC results
self.add_step(Step('run-search-exp', self.run, stage='search'))
self.add_fetcher(src=self.search_exp_path, name='fetch-search-results')
def __init__(
self,
suites,
num_runs=30,
time_per_step=1.0,
initial_port=2000,
rddlsim_seed=0,
rddlsim_enforces_runtime=False,
revision_cache=None,
time_buffer=300,
memory_limit=int(3.5 * 1024),
soft_stdout_limit=10 * 1024,
hard_stdout_limit=20 * 1024,
soft_stderr_limit=64,
hard_stderr_limit=10 * 1024,
path=None,
environment=None,
):
"""
*suites* is a list of :class:'prostlab.suites.Problem' objects that describes
the set of benchmarks used in the experiment.
*num_runs* is the number of times each algorithm is executed on each instance.
*time_per_step* is the time in seconds each algorithm has per step. A total time
per instance is computed from this, the *num_runs*, the instance horizon and the
*time_buffer*.
*initial_port* is the first port that is used for TCP/IP communication between
an algorithm and rddlsim.
*rddlsim_seed* is the value with which rddlsim is seeded.
If *rddlsim_enforces_runtime* is True, rddlsim terminates after the time that is
computed as the product of *num_runs*, *time_per_step* and the instance horizon.
*revision_cache* is the directory for caching Prost revisions. It defaults to
``<scriptdir>/data/revision-cache``.
*time_buffer* is the time that is allows in addtion to the product of *num_runs*,
*time_per_step* and the instance horizon. This must include the time the parser
requires.
*memory_limit* is the hard memory limit in MiB. *memory_limit* - 512 MiB is
furthermore passed as a (soft) memory limit to Prost.
*soft_stdout_limit*, *hard_stdout_limit*, *soft_stderr_limit* and
*hard_stderr_limit* limit the amount of data each experiment may write to disk,
See :class:`lab.experiment.Experiment` for an explanation of
the *path* and *environment* parameters.
>>> from lab.environments import BaselSlurmEnvironment
>>> env = BaselSlurmEnvironment(email="*****@*****.**")
>>> exp = ProstExperiment(environment=env)
You can add parsers with :meth:`.add_parser()`. See
:ref:`parsing` for how to write custom parsers.
"""
Experiment.__init__(self, path=path, environment=environment)
self.suites = suites
self.num_runs = num_runs
self.time_per_step = time_per_step
self.initial_port = initial_port
self.rddlsim_seed = rddlsim_seed
self.rddlsim_enforces_runtime = rddlsim_enforces_runtime
self.revision_cache = revision_cache or os.path.join(
get_default_data_dir(), "revision-cache")
self.time_buffer = time_buffer
self.memory_limit = memory_limit
self.soft_stdout_limit = soft_stdout_limit
self.hard_stdout_limit = hard_stdout_limit
self.soft_stderr_limit = soft_stderr_limit
self.hard_stderr_limit = hard_stderr_limit
# Use OrderedDict to ensure that names are unique and ordered.
self.configs = OrderedDict()
def __init__(self, path, repo, environment=None, combinations=None,
compact=True, limits=None, cache_dir=None):
"""
The experiment will be built at *path*.
*repo* must be the path to a Fast Downward repository. Among other things
this repository is used to search for benchmark files.
*environment* must be an :ref:`Environment <environments>` instance.
By default the experiment is run locally.
If given, *combinations* must be a list of :ref:`Checkout <checkouts>`
tuples of the form (Translator, Preprocessor, Planner). If combinations
is None (default), perform an experiment with the working copy in *repo*.
The *compact* parameter is only relevant for the search
stage. If *compact* is ``False``, the preprocessed task and
the two PDDL files are **copied** into the respective run
directories for all configurations. This requires a lot of
space (tens of GB), so it is strongly recommended to use the
default (``compact=True``) which only references these
files. Use ``compact=False`` only if you really need a
portable experiment.
If *limits* is given, it must be a dictionary that maps a
subset of the keys below to seconds and MiB. It will be used
to overwrite the default limits::
default_limits = {
'translate_time': 7200,
'translate_memory': 8192,
'preprocess_time': 7200,
'preprocess_memory': 8192,
'search_time': 1800,
'search_memory': 2048,
}
*cache_dir* is used to cache Fast Downward clones and preprocessed
tasks. By default it points to ``~/lab``.
.. note::
The directory *cache_dir* can grow very large (tens of GB).
Example: ::
repo = '/path/to/downward-repo'
env = GkiGridEnvironment(queue='xeon_core.q', priority=-2)
combos = [(Translator(repo, rev=123),
Preprocessor(repo, rev='e2a018c865f7'),
Planner(repo, rev='tip')]
exp = DownwardExperiment('/tmp/path', repo, environment=env,
combinations=combos,
limits={'search_time': 30,
'search_memory': 1024})
"""
Experiment.__init__(self, path, environment=environment, cache_dir=cache_dir)
if not repo or not os.path.isdir(repo):
logging.critical('The path "%s" is not a local Fast Downward '
'repository.' % repo)
self.repo = repo
self.orig_path = self.path
self.search_exp_path = self.path
self.preprocess_exp_path = self.path + '-p'
self._path_to_python = None
Checkout.REV_CACHE_DIR = os.path.join(self.cache_dir, 'revision-cache')
self.preprocessed_tasks_dir = os.path.join(self.cache_dir, 'preprocessed-tasks')
tools.makedirs(self.preprocessed_tasks_dir)
self.combinations = (combinations or
[(Translator(repo), Preprocessor(repo), Planner(repo))])
self.compact = compact
self.suites = defaultdict(list)
self._algorithms = []
self._portfolios = []
limits = limits or {}
for key, value in limits.items():
if key not in LIMITS:
logging.critical('Unknown limit: %s' % key)
self.limits = LIMITS
self.limits.update(limits)
# Save if this is a compact experiment i.e. preprocessed tasks are referenced.
self.set_property('compact', compact)
# TODO: Integrate this into the API.
self.include_preprocess_results_in_search_runs = True
self.compilation_options = ['-j%d' % self._jobs]
self._search_parsers = []
self.add_search_parser(os.path.join(DOWNWARD_SCRIPTS_DIR, 'search_parser.py'))
# Remove the default experiment steps
self.steps = Sequence()
self.add_step(Step('build-preprocess-exp', self.build, stage='preprocess'))
self.add_step(Step('run-preprocess-exp', self.run, stage='preprocess'))
self.add_fetcher(src=self.preprocess_exp_path,
dest=self.preprocessed_tasks_dir,
name='fetch-preprocess-results',
copy_all=True,
write_combined_props=False)
self.add_step(Step('build-search-exp', self.build, stage='search'))
self.add_PAC_fetcher(src='/home/gal-d/downward/lab/examples/PAC_Preprocess_Output-eval/preprocess',#TODO change to be parameter
dest=self.search_exp_path,
name='fetch-preprocess-results',
copy_all=True,
write_combined_props=False)#new featcher to copy preprocess for PAC results
self.add_step(Step('run-search-exp', self.run, stage='search'))
self.add_fetcher(src=self.search_exp_path, name='fetch-search-results')
