""" Option parser for bumps command line """ from __future__ import print_function import sys import numpy as np from .fitters import FITTERS, FIT_AVAILABLE_IDS, FIT_ACTIVE_IDS, FIT_DEFAULT_ID # TODO: replace with standard argparse module class ParseOpts(object): """ Options parser. Subclass should define *MINARGS*, *FLAGS*, *VALUES* and *USAGE*. *MINARGS* is the minimum number of positional arguments. *FLAGS* is a set of arguments that may be present or absent. *VALUES* is a set of arguments that take values. Value checking can be done in the setter for each argument in the set. Default values should be set in the corresponding object attribute. *USAGE* is the help string to display for option "help". The constructor will invoke the command line parser, leaving the values set by the command line as attribute values. Flag options will be True or False. """ MINARGS = 0 FLAGS = set() VALUES = set() #: Value to use if a value flag is is present without '='. This is #: different from the default value if the flag is not present, which #: is the default value set in the calling class. IMPLICIT_VALUES = {} USAGE = "" def __init__(self, args): if self.VALUES & self.FLAGS: raise TypeError("option used as both a flag and a value: %s"% ",".join(self.VALUES&self.FLAGS)) self._parse(args) def _parse(self, args): if any(v in args for v in ('-?', '-h', '-help')): print(self.USAGE) sys.exit() # Drop the "bumps" arg from the beginning of the list. args = args[1:] # Fill in implicit values. We need to do this to support something # like "bumps ... --parallel=2 ... --parallel", which should have the # later (implicit) parameter take precedence over the earlier # parameters. args = [ (arg + "=" + str(self.IMPLICIT_VALUES[arg[2:]]) if arg[2:] in self.IMPLICIT_VALUES else arg) for arg in args] # Parse options. # Given tuples [..., (a, 1), ..., (a, 2), ...], then dict(tuples) # will use the later value for the key rather than the earlier # value, which is what we want for the command line interpreter. position_args = [v for v in sys.argv[1:] if not v.startswith('--')] flag_args = [ v[2:] # convert --flag => flag for v in args if v.startswith('--') and not '=' in v] value_args = dict( v[2:].split('=', 1) # convert --flag=value => (flag, value) for v in args if v.startswith('--') and '=' in v) # Check that options are valid. # TODO: move type checking from FitConfig.set_from_cli to here. flags, values = set(flag_args), set(value_args.keys()) unknown = (flags|values) - (self.FLAGS|self.VALUES) unexpected_value = flags - self.FLAGS blank_values = set( k for k, v in value_args.items() if k in self.VALUES and v == "") missing_value = (values - self.VALUES) | blank_values errors = [] if any(unknown): errors.append( "Unknown options --%s." % ", --".join(unknown)) if any(unexpected_value): errors.append( "Unexpected value for --%s." % ", --".join(sorted(unexpected_value))) if any(missing_value): errors.append( "Missing value for --%s." % ", --".join(sorted(missing_value))) if errors: message = " ".join(errors + ["Use -? for help."]) raise ValueError(message) # Set the values into the fields. for option in self.FLAGS: setattr(self, option, (option in flags)) for option, value in value_args.items(): setattr(self, option, value) self.args = position_args # === Fitter option parsing === class ChoiceList(object): def __init__(self, *choices): self.choices = choices def __call__(self, value): if not value in self.choices: raise ValueError('invalid option "%s": use %s' % (value, '|'.join(self.choices))) else: return value def yesno(value): if value.lower() in ('true', 'yes', 'on', '1'): return True elif value.lower() in ('false', 'no', 'off', '0'): return False raise ValueError('invalid option "%s": use yes|no') def parse_int(value): float_value = float(value) if int(float_value) != float_value: raise ValueError("integer expected") return int(float_value) FIT_FIELDS = dict( starts=("Starts", parse_int), steps=("Steps", parse_int), samples=("Samples", parse_int), xtol=("x tolerance", float), ftol=("f(x) tolerance", float), alpha=("Convergence", float), stop=("Stopping criteria", str), thin=("Thinning", parse_int), burn=("Burn-in steps", parse_int), pop=("Population", float), init=("Initializer", ChoiceList("eps", "lhs", "cov", "random")), CR=("Crossover ratio", float), F=("Scale", float), nT=("# Temperatures", parse_int), Tmin=("Min temperature", float), Tmax=("Max temperature", float), radius=("Simplex radius", float), # TODO: convert --trim into a boolean flag and update docs trim=("Burn-in trim", yesno), outliers=("Outliers" , ChoiceList("none", "iqr", "grubbs", "mahal")), ) # Make sure all settings are parseable for fit in FITTERS: assert all(opt in FIT_FIELDS for opt, _ in fit.settings), ( "Fitter %s contains unknown settings %s" %(fit.id, ', '.join(opt for opt, _ in sorted(fit.settings) if opt not in FIT_FIELDS))) del fit class FitConfig(object): """ Fit settings configuration object. The command line parser will define a FitConfig object which contains the fitter that was given on the command line and all its options. For embedded bumps, which does not use the bumps command line parser, a new FitConfig object can be created with its own selected options. **Attributes** *ids = [id, id, ...]* is a list available fitters in "preferred" order. Depending on usage, you may want to sort them, or alternatively, sort by long name with *[id for _,id in sorted((v,k) for k,v in self.names]* *fitters = {id: fitclasss}* maps ids to fitters. *names* = {id: name}* maps ids to long names *settings = {id: [(option, default), ...]}* maps ids to default settings. The order of the settings is the preferred order to present the settings to the user in a GUI dialog for example. *values = {id: {option: value, ...}}* maps ids to the settings for each fitter. Note that in the GUI, different fitters may have there settings recorded and preserved even when not selected. *active_ids = [id, id, ...]* is the list of fitters to show the user in a GUI dialog for example. The other fitters should still be available from the command line. *default_id = id* is the fitter to use by default. *selected_id = id* is the fitter that was selected, either by command line or by GUI. *selected_values = {option: value}* returns the settings for the current fitter. *selected_name = name* returns the name of the selected fitter. *selected_fitter = FitClass* returns the class of the selected fitter. """ def __init__(self, default=FIT_DEFAULT_ID, active=FIT_ACTIVE_IDS): # Keep a private copy of the configure settings rather than modifying # the global defaults self.ids = [fit.id for fit in FITTERS] # FITTERS is a list of FitBase classes # Each class has: # fit.id: the short name used on the command line # fit.name: the long name used in the GUI # fit.settings: available options: [(key,default value), ...] self.fitters = dict((fit.id, fit) for fit in FITTERS) self.names = dict((fit.id, fit.name) for fit in FITTERS) self.settings = dict((fit.id, fit.settings) for fit in FITTERS) self.values = dict((fit.id, dict(fit.settings)) for fit in FITTERS) if not all(k in self.ids for k in active): raise ValueError("Some active fitters are not available") if default not in active: raise ValueError("default fitter is not active") self.active_ids = active self.default_id = default self.selected_id = default def set_from_cli(self, opts): """ Use the BumpsOpts command line parser values to set the selected fitter and its configuration options. """ fitter = opts.fit self.selected_id = fitter # Convert supplied options to the correct types and save them in value for field, reset_value in self.settings[fitter]: value = getattr(opts, field, None) parse = FIT_FIELDS[field][1] if value is not None: try: self.values[fitter][field] = parse(value) except Exception as exc: raise ValueError("error in --%s: %s" % (field, str(exc))) # print("options=%s"%(str(self.options))) @property def selected_values(self): return self.values[self.selected_id] @property def selected_name(self): return self.names[self.selected_id] @property def selected_fitter(self): return self.fitters[self.selected_id] #: FitConfig singleton for the common case in which only one config is needed. #: There may be other use cases, such as saving the fit config along with the #: rest of the state so that on resume the fit options are restored, but in that #: case the application will not be using the singleton. FIT_CONFIG = FitConfig() # === Bumps options parsing === class BumpsOpts(ParseOpts): """ Option parser for bumps. """ MINARGS = 1 # TODO: document all options in USAGE and doc/guide/options.rst # TODO: remove application-specific options like --staj FLAGS = set(("preview", "chisq", "profile", "time_model", "simulate", "simrandom", "shake", "worker", # internal, so not documented "multiprocessing-fork", # passed in when app is a frozen image "remote", # not active, so not documented "batch", "noshow", "overwrite", "stepmon", "err", "cov", "edit", "mpi", "keep_best", "staj", # passed when not running bumps, but instead using a # bundled application as a python distribution with domain # specific models pre-defined. "i", )) VALUES = set(("plot", "store", "resume", "entropy", "fit", "noise", "seed", "pars", "resynth", "time", "checkpoint", "m", "c", "p", "parallel", "view", "trim", "alpha", "outliers", # The following options are for remote fitting via the # fitting service, but this is not currently active. "transport", "notify", "queue", )) # Add in parameters from the fitters VALUES |= set(FIT_FIELDS.keys()) # --parallel is equivalent to --parallel=0 IMPLICIT_VALUES = { 'parallel': '0', 'entropy': 'llf', 'resume': '-', } pars = None notify = "" queue = "http://reflectometry.org/queue" resynth = "0" noise = "5" starts = "1" seed = "" time = "inf" checkpoint = "0" parallel = "" entropy = None trim = "true" view = None alpha = 0.0 PLOTTERS = "linear", "log", "residuals" USAGE = """\ Usage: bumps [options] modelfile [modelargs] The modelfile is a Python script (i.e., a series of Python commands) which sets up the data, the models, and the fittable parameters. The model arguments are available in the modelfile as sys.argv[1:]. Model arguments may not start with '-'. Options: --preview display model but do not perform a fitting operation --pars=filename or store path initial parameter values; fit results are saved as path/.par --plot=log [%(plotter)s] type of plot to display --trim=true trim any remaining burn before displaying plots [dream only] --simulate simulate a dataset using the initial problem parameters --simrandom simulate a dataset using random problem parameters --shake set random parameters before fitting --noise=5%% percent noise to add to the simulated data --seed=integer random number seed --err show uncertainty estimate from curvature at the minimum --cov show the covariance matrix for the model when done --entropy=gmm|mvn|wnn|llf compute entropy on posterior distribution [dream only] --staj output staj file when done [Refl1D only] --edit start the gui --view=linear|log one of the predefined problem views; reflectometry also has fresnel, logfresnel, q4 and residuals --store=path output directory for plots and models --overwrite if store already exists, replace it --resume=path [dream] resume a fit from previous stored state; if path is '-' then use the path given by --store, if it exists --parallel=n run fit using multiprocessing for parallelism; use --parallel=0 for all cpus --mpi run fit using MPI for parallelism (use command "mpirun -n cpus ...") --batch batch mode; save output in .mon file and don't show plots after fit --noshow semi-batch; send output to console but don't show plots after fit --remote queue fit to run on remote server --notify=user@email remote fit notification --queue=http://reflectometry.org remote job queue --time=inf run for a maximum number of hours --checkpoint=0 save fit state every n hours, or 0 for no checkpoints --fit=amoeba [%(fitter)s] fitting engine to use; see manual for details --steps=0 [%(fitter)s] number of fit iterations after any burn-in time; use samples if steps=0 --samples=1e4 [dream] set steps=samples/(pop*#pars) so the target number of samples is drawn --xtol=1e-4 [de, amoeba] minimum population diameter --ftol=1e-4 [de, amoeba] minimum population flatness --alpha=0.0 [dream] p-level for rejecting convergence; with fewer samples use a stricter stopping condition, such as --alpha=0.01 --samples=20000 --outliers=none [dream] name of test used for removing outlier chains every N samples: none: no outlier removal iqr: use interquartile range on likelihood grubbs: use t-test on likelihood mahal: use distance from parameter values on the best chain --pop=10 [dream, de, rl, ps] population size is pop times number of fitted parameters; if pop is negative, then set population size to -pop. --burn=100 [dream, pt] number of burn-in iterations before accumulating stats --thin=1 [dream] number of fit iterations between steps --nT=25 --Tmin=0.1 --Tmax=10 [pt] temperatures vector; use a higher maximum temperature and a larger nT if your fit is getting stuck in local minima --CR=0.9 [de, rl, pt] crossover ratio for population mixing --starts=1 [newton, rl, amoeba] number of times to run the fit from random starting points. --keep_best when running with multiple starts, restart from a point near the last minimum rather than using a completely random starting point. --init=eps [dream] population initialization method: eps: ball around initial parameter set lhs: latin hypercube sampling cov: normally distributed according to covariance matrix random: uniformly distributed within parameter ranges --stepmon show details for each step --resynth=0 run resynthesis error analysis for n generations --time_model run the model --steps times in order to estimate total run time. --profile run the python profiler on the model; use --steps to run multiple models for better statistics --chisq print the model description and chisq value and exit -m/-c/-p command run the python interpreter with bumps on the path: m: command is a module such as bumps.cli, run as __main__ c: command is a python one-line command p: command is the name of a python script -i start the interactive interpreter -?/-h/--help display this help """ % {'fitter': '|'.join(sorted(FIT_AVAILABLE_IDS)), 'plotter': '|'.join(PLOTTERS), } # --transport=mp {amqp|mp|mpi} # use amqp/multiprocessing/mpi for parallel evaluation _plot = 'log' def _set_plot(self, value): if value not in set(self.PLOTTERS): raise ValueError("unknown plot type %s; use %s" % (value, "|".join(self.PLOTTERS))) self._plot = value plot = property(fget=lambda self: self._plot, fset=_set_plot) store = None resume = None _fit = FIT_DEFAULT_ID @property def fit(self): return self._fit @fit.setter def fit(self, value): if value not in FIT_AVAILABLE_IDS: raise ValueError("unknown fitter %s; use %s" % (value, "|".join(sorted(FIT_AVAILABLE_IDS)))) self._fit = value fit_config = FIT_CONFIG TRANSPORTS = 'amqp', 'mp', 'mpi', 'celery' _transport = 'mp' def _set_transport(self, value): if value not in self.TRANSPORTS: raise ValueError("unknown transport %s; use %s" % (value, "|".join(self.TRANSPORTS))) self._transport = value transport = property(fget=lambda self: self._transport, fset=_set_transport) meshsteps = 40 def getopts(): """ Process command line options. Option values will be stored as attributes in the returned object. """ opts = BumpsOpts(sys.argv) opts.resynth = int(opts.resynth) # Set a random seed if none is given; want to know the seed so we can # reproduce the run. The seed needs to be saved to the monitor file. opts.seed = int(opts.seed) if opts.seed else np.random.randint(1000000) opts.fit_config.set_from_cli(opts) return opts