""" Container for the result of running the sample (MCMC) method """ import math import os from io import StringIO from typing import ( Any, Dict, Hashable, List, MutableMapping, Optional, Sequence, Tuple, Union, ) import numpy as np import pandas as pd try: import xarray as xr XARRAY_INSTALLED = True except ImportError: XARRAY_INSTALLED = False from cmdstanpy import _CMDSTAN_SAMPLING, _CMDSTAN_THIN, _CMDSTAN_WARMUP, _TMPDIR from cmdstanpy.cmdstan_args import Method, SamplerArgs from cmdstanpy.utils import ( EXTENSION, build_xarray_data, check_sampler_csv, cmdstan_path, cmdstan_version_before, create_named_text_file, do_command, flatten_chains, get_logger, ) from .metadata import InferenceMetadata from .runset import RunSet class CmdStanMCMC: """ Container for outputs from CmdStan sampler run. Provides methods to summarize and diagnose the model fit and accessor methods to access the entire sample or individual items. Created by :meth:`CmdStanModel.sample` The sample is lazily instantiated on first access of either the resulting sample or the HMC tuning parameters, i.e., the step size and metric. """ # pylint: disable=too-many-public-methods def __init__( self, runset: RunSet, ) -> None: """Initialize object.""" if not runset.method == Method.SAMPLE: raise ValueError( 'Wrong runset method, expecting sample runset, ' 'found method {}'.format(runset.method) ) self.runset = runset # info from runset to be exposed sampler_args = self.runset._args.method_args assert isinstance( sampler_args, SamplerArgs ) # make the typechecker happy self._iter_sampling: int = _CMDSTAN_SAMPLING if sampler_args.iter_sampling is not None: self._iter_sampling = sampler_args.iter_sampling self._iter_warmup: int = _CMDSTAN_WARMUP if sampler_args.iter_warmup is not None: self._iter_warmup = sampler_args.iter_warmup self._thin: int = _CMDSTAN_THIN if sampler_args.thin is not None: self._thin = sampler_args.thin self._is_fixed_param = sampler_args.fixed_param self._save_warmup: bool = sampler_args.save_warmup self._sig_figs = runset._args.sig_figs # info from CSV values, instantiated lazily self._draws: np.ndarray = np.array(()) # only valid when not is_fixed_param self._metric: np.ndarray = np.array(()) self._step_size: np.ndarray = np.array(()) self._divergences: np.ndarray = np.zeros(self.runset.chains, dtype=int) self._max_treedepths: np.ndarray = np.zeros( self.runset.chains, dtype=int ) # info from CSV initial comments and header config = self._validate_csv_files() self._metadata: InferenceMetadata = InferenceMetadata(config) if not self._is_fixed_param: self._check_sampler_diagnostics() def __repr__(self) -> str: repr = 'CmdStanMCMC: model={} chains={}{}'.format( self.runset.model, self.runset.chains, self.runset._args.method_args.compose(0, cmd=[]), ) repr = '{}\n csv_files:\n\t{}\n output_files:\n\t{}'.format( repr, '\n\t'.join(self.runset.csv_files), '\n\t'.join(self.runset.stdout_files), ) # TODO - hamiltonian, profiling files return repr def __getattr__(self, attr: str) -> np.ndarray: """Synonymous with ``fit.stan_variable(attr)""" if attr.startswith("_"): raise AttributeError(f"Unknown variable name {attr}") try: return self.stan_variable(attr) except ValueError as e: # pylint: disable=raise-missing-from raise AttributeError(*e.args) def __getstate__(self) -> dict: # This function returns the mapping of objects to serialize with pickle. # See https://docs.python.org/3/library/pickle.html#object.__getstate__ # for details. We call _assemble_draws to ensure posterior samples have # been loaded prior to serialization. self._assemble_draws() return self.__dict__ @property def chains(self) -> int: """Number of chains.""" return self.runset.chains @property def chain_ids(self) -> List[int]: """Chain ids.""" return self.runset.chain_ids @property def num_draws_warmup(self) -> int: """Number of warmup draws per chain, i.e., thinned warmup iterations.""" return int(math.ceil((self._iter_warmup) / self._thin)) @property def num_draws_sampling(self) -> int: """ Number of sampling (post-warmup) draws per chain, i.e., thinned sampling iterations. """ return int(math.ceil((self._iter_sampling) / self._thin)) @property def metadata(self) -> InferenceMetadata: """ Returns object which contains CmdStan configuration as well as information about the names and structure of the inference method and model output variables. """ return self._metadata @property def column_names(self) -> Tuple[str, ...]: """ Names of all outputs from the sampler, comprising sampler parameters and all components of all model parameters, transformed parameters, and quantities of interest. Corresponds to Stan CSV file header row, with names munged to array notation, e.g. `beta[1]` not `beta.1`. """ return self._metadata.cmdstan_config['column_names'] # type: ignore @property def metric_type(self) -> Optional[str]: """ Metric type used for adaptation, either 'diag_e' or 'dense_e', according to CmdStan arg 'metric'. When sampler algorithm 'fixed_param' is specified, metric_type is None. """ return ( self._metadata.cmdstan_config['metric'] if not self._is_fixed_param else None ) @property def metric(self) -> Optional[np.ndarray]: """ Metric used by sampler for each chain. When sampler algorithm 'fixed_param' is specified, metric is None. """ if self._is_fixed_param: return None if self._metadata.cmdstan_config['metric'] == 'unit_e': get_logger().info( 'Unit diagnonal metric, inverse mass matrix size unknown.' ) return None self._assemble_draws() return self._metric @property def step_size(self) -> Optional[np.ndarray]: """ Step size used by sampler for each chain. When sampler algorithm 'fixed_param' is specified, step size is None. """ self._assemble_draws() return self._step_size if not self._is_fixed_param else None @property def thin(self) -> int: """ Period between recorded iterations. (Default is 1). """ return self._thin @property def divergences(self) -> Optional[np.ndarray]: """ Per-chain total number of post-warmup divergent iterations. When sampler algorithm 'fixed_param' is specified, returns None. """ return self._divergences if not self._is_fixed_param else None @property def max_treedepths(self) -> Optional[np.ndarray]: """ Per-chain total number of post-warmup iterations where the NUTS sampler reached the maximum allowed treedepth. When sampler algorithm 'fixed_param' is specified, returns None. """ return self._max_treedepths if not self._is_fixed_param else None def draws( self, *, inc_warmup: bool = False, concat_chains: bool = False ) -> np.ndarray: """ Returns a numpy.ndarray over all draws from all chains which is stored column major so that the values for a parameter are contiguous in memory, likewise all draws from a chain are contiguous. By default, returns a 3D array arranged (draws, chains, columns); parameter ``concat_chains=True`` will return a 2D array where all chains are flattened into a single column, preserving chain order, so that given M chains of N draws, the first N draws are from chain 1, up through the last N draws from chain M. :param inc_warmup: When ``True`` and the warmup draws are present in the output, i.e., the sampler was run with ``save_warmup=True``, then the warmup draws are included. Default value is ``False``. :param concat_chains: When ``True`` return a 2D array flattening all all draws from all chains. Default value is ``False``. See Also -------- CmdStanMCMC.draws_pd CmdStanMCMC.draws_xr CmdStanGQ.draws """ self._assemble_draws() if inc_warmup and not self._save_warmup: get_logger().warning( "Sample doesn't contain draws from warmup iterations," ' rerun sampler with "save_warmup=True".' ) start_idx = 0 if not inc_warmup and self._save_warmup: start_idx = self.num_draws_warmup if concat_chains: return flatten_chains(self._draws[start_idx:, :, :]) return self._draws[start_idx:, :, :] def _validate_csv_files(self) -> Dict[str, Any]: """ Checks that Stan CSV output files for all chains are consistent and returns dict containing config and column names. Tabulates sampling iters which are divergent or at max treedepth Raises exception when inconsistencies detected. """ dzero = {} for i in range(self.chains): if i == 0: dzero = check_sampler_csv( path=self.runset.csv_files[i], is_fixed_param=self._is_fixed_param, iter_sampling=self._iter_sampling, iter_warmup=self._iter_warmup, save_warmup=self._save_warmup, thin=self._thin, ) if not self._is_fixed_param: self._divergences[i] = dzero['ct_divergences'] self._max_treedepths[i] = dzero['ct_max_treedepth'] else: drest = check_sampler_csv( path=self.runset.csv_files[i], is_fixed_param=self._is_fixed_param, iter_sampling=self._iter_sampling, iter_warmup=self._iter_warmup, save_warmup=self._save_warmup, thin=self._thin, ) for key in dzero: # check args that matter for parsing, plus name, version if ( key in [ 'stan_version_major', 'stan_version_minor', 'stan_version_patch', 'stanc_version', 'model', 'num_samples', 'num_warmup', 'save_warmup', 'thin', 'refresh', ] and dzero[key] != drest[key] ): raise ValueError( 'CmdStan config mismatch in Stan CSV file {}: ' 'arg {} is {}, expected {}'.format( self.runset.csv_files[i], key, dzero[key], drest[key], ) ) if not self._is_fixed_param: self._divergences[i] = drest['ct_divergences'] self._max_treedepths[i] = drest['ct_max_treedepth'] return dzero def _check_sampler_diagnostics(self) -> None: """ Warn if any iterations ended in divergences or hit maxtreedepth. """ if np.any(self._divergences) or np.any(self._max_treedepths): diagnostics = ['Some chains may have failed to converge.'] ct_iters = self._metadata.cmdstan_config['num_samples'] for i in range(self.runset._chains): if self._divergences[i] > 0: diagnostics.append( f'Chain {i + 1} had {self._divergences[i]} ' 'divergent transitions ' f'({((self._divergences[i]/ct_iters)*100):.1f}%)' ) if self._max_treedepths[i] > 0: diagnostics.append( f'Chain {i + 1} had {self._max_treedepths[i]} ' 'iterations at max treedepth ' f'({((self._max_treedepths[i]/ct_iters)*100):.1f}%)' ) diagnostics.append( 'Use the "diagnose()" method on the CmdStanMCMC object' ' to see further information.' ) get_logger().warning('\n\t'.join(diagnostics)) def _assemble_draws(self) -> None: """ Allocates and populates the step size, metric, and sample arrays by parsing the validated stan_csv files. """ if self._draws.shape != (0,): return num_draws = self.num_draws_sampling sampling_iter_start = 0 if self._save_warmup: num_draws += self.num_draws_warmup sampling_iter_start = self.num_draws_warmup self._draws = np.empty( (num_draws, self.chains, len(self.column_names)), dtype=float, order='F', ) self._step_size = np.empty(self.chains, dtype=float) for chain in range(self.chains): with open(self.runset.csv_files[chain], 'r') as fd: line = fd.readline().strip() # read initial comments, CSV header row while len(line) > 0 and line.startswith('#'): line = fd.readline().strip() if not self._is_fixed_param: # handle warmup draws, if any if self._save_warmup: for i in range(self.num_draws_warmup): line = fd.readline().strip() xs = line.split(',') self._draws[i, chain, :] = [float(x) for x in xs] line = fd.readline().strip() if line != '# Adaptation terminated': # shouldn't happen? while line != '# Adaptation terminated': line = fd.readline().strip() # step_size, metric (diag_e and dense_e only) line = fd.readline().strip() _, step_size = line.split('=') self._step_size[chain] = float(step_size.strip()) if self._metadata.cmdstan_config['metric'] != 'unit_e': line = fd.readline().strip() # metric type line = fd.readline().lstrip(' #\t').rstrip() num_unconstrained_params = len(line.split(',')) if chain == 0: # can't allocate w/o num params if self.metric_type == 'diag_e': self._metric = np.empty( (self.chains, num_unconstrained_params), dtype=float, ) else: self._metric = np.empty( ( self.chains, num_unconstrained_params, num_unconstrained_params, ), dtype=float, ) if line: if self.metric_type == 'diag_e': xs = line.split(',') self._metric[chain, :] = [float(x) for x in xs] else: xs = line.strip().split(',') self._metric[chain, 0, :] = [ float(x) for x in xs ] for i in range(1, num_unconstrained_params): line = fd.readline().lstrip(' #\t').rstrip() xs = line.split(',') self._metric[chain, i, :] = [ float(x) for x in xs ] else: # unit_e changed in 2.34 to have an extra line pos = fd.tell() line = fd.readline().strip() if not line.startswith('#'): fd.seek(pos) # process draws for i in range(sampling_iter_start, num_draws): line = fd.readline().strip() xs = line.split(',') self._draws[i, chain, :] = [float(x) for x in xs] assert self._draws is not None def summary( self, percentiles: Sequence[int] = (5, 50, 95), sig_figs: int = 6, ) -> pd.DataFrame: """ Run cmdstan/bin/stansummary over all output CSV files, assemble summary into DataFrame object. The first row contains statistics for the total joint log probability `lp__`, but is omitted when the Stan model has no parameters. The remaining rows contain summary statistics for all parameters, transformed parameters, and generated quantities variables, in program declaration order. :param percentiles: Ordered non-empty sequence of percentiles to report. Must be integers from (1, 99), inclusive. Defaults to ``(5, 50, 95)`` :param sig_figs: Number of significant figures to report. Must be an integer between 1 and 18. If unspecified, the default precision for the system file I/O is used; the usual value is 6. If precision above 6 is requested, sample must have been produced by CmdStan version 2.25 or later and sampler output precision must equal to or greater than the requested summary precision. :return: pandas.DataFrame """ if len(percentiles) == 0: raise ValueError( 'Invalid percentiles argument, must be ordered' ' non-empty list from (1, 99), inclusive.' ) cur_pct = 0 for pct in percentiles: if pct > 99 or not pct > cur_pct: raise ValueError( 'Invalid percentiles spec, must be ordered' ' non-empty list from (1, 99), inclusive.' ) cur_pct = pct percentiles_str = ( f"--percentiles= {','.join(str(x) for x in percentiles)}" ) if not isinstance(sig_figs, int) or sig_figs < 1 or sig_figs > 18: raise ValueError( 'Keyword "sig_figs" must be an integer between 1 and 18,' ' found {}'.format(sig_figs) ) csv_sig_figs = self._sig_figs or 6 if sig_figs > csv_sig_figs: get_logger().warning( 'Requesting %d significant digits of output, but CSV files' ' only have %d digits of precision.', sig_figs, csv_sig_figs, ) sig_figs_str = f'--sig_figs={sig_figs}' cmd_path = os.path.join( cmdstan_path(), 'bin', 'stansummary' + EXTENSION ) tmp_csv_file = 'stansummary-{}-'.format(self.runset._args.model_name) tmp_csv_path = create_named_text_file( dir=_TMPDIR, prefix=tmp_csv_file, suffix='.csv', name_only=True ) csv_str = '--csv_filename={}'.format(tmp_csv_path) # TODO: remove at some future release if cmdstan_version_before(2, 24): csv_str = '--csv_file={}'.format(tmp_csv_path) cmd = [ cmd_path, percentiles_str, sig_figs_str, csv_str, ] + self.runset.csv_files do_command(cmd, fd_out=None) with open(tmp_csv_path, 'rb') as fd: summary_data = pd.read_csv( fd, delimiter=',', header=0, index_col=0, comment='#', float_precision='high', ) mask = ( [not x.endswith('__') for x in summary_data.index] if self._is_fixed_param else [ x == 'lp__' or not x.endswith('__') for x in summary_data.index ] ) summary_data.index.name = None return summary_data[mask] def diagnose(self) -> Optional[str]: """ Run cmdstan/bin/diagnose over all output CSV files, return console output. The diagnose utility reads the outputs of all chains and checks for the following potential problems: + Transitions that hit the maximum treedepth + Divergent transitions + Low E-BFMI values (sampler transitions HMC potential energy) + Low effective sample sizes + High R-hat values """ cmd_path = os.path.join(cmdstan_path(), 'bin', 'diagnose' + EXTENSION) cmd = [cmd_path] + self.runset.csv_files result = StringIO() do_command(cmd=cmd, fd_out=result) return result.getvalue() def draws_pd( self, vars: Union[List[str], str, None] = None, inc_warmup: bool = False, ) -> pd.DataFrame: """ Returns the sample draws as a pandas DataFrame. Flattens all chains into single column. Container variables (array, vector, matrix) will span multiple columns, one column per element. E.g. variable 'matrix[2,2] foo' spans 4 columns: 'foo[1,1], ... foo[2,2]'. :param vars: optional list of variable names. :param inc_warmup: When ``True`` and the warmup draws are present in the output, i.e., the sampler was run with ``save_warmup=True``, then the warmup draws are included. Default value is ``False``. See Also -------- CmdStanMCMC.draws CmdStanMCMC.draws_xr CmdStanGQ.draws_pd """ if vars is not None: if isinstance(vars, str): vars_list = [vars] else: vars_list = vars if inc_warmup and not self._save_warmup: get_logger().warning( 'Draws from warmup iterations not available,' ' must run sampler with "save_warmup=True".' ) self._assemble_draws() cols = [] if vars is not None: for var in dict.fromkeys(vars_list): if var in self._metadata.method_vars: cols.append(var) elif var in self._metadata.stan_vars: info = self._metadata.stan_vars[var] cols.extend( self.column_names[info.start_idx : info.end_idx] ) elif var in ['chain__', 'iter__', 'draw__']: cols.append(var) else: raise ValueError(f'Unknown variable: {var}') else: cols = ['chain__', 'iter__', 'draw__'] + list(self.column_names) draws = self.draws(inc_warmup=inc_warmup) # add long-form columns for chain, iteration, draw n_draws, n_chains, _ = draws.shape chains_col = ( np.repeat(np.arange(1, n_chains + 1), n_draws) .reshape(1, n_chains, n_draws) .T ) iter_col = ( np.tile(np.arange(1, n_draws + 1), n_chains) .reshape(1, n_chains, n_draws) .T ) draw_col = ( np.arange(1, (n_draws * n_chains) + 1) .reshape(1, n_chains, n_draws) .T ) draws = np.concatenate([chains_col, iter_col, draw_col, draws], axis=2) return pd.DataFrame( data=flatten_chains(draws), columns=['chain__', 'iter__', 'draw__'] + list(self.column_names), )[cols] def draws_xr( self, vars: Union[str, List[str], None] = None, inc_warmup: bool = False ) -> "xr.Dataset": """ Returns the sampler draws as a xarray Dataset. :param vars: optional list of variable names. :param inc_warmup: When ``True`` and the warmup draws are present in the output, i.e., the sampler was run with ``save_warmup=True``, then the warmup draws are included. Default value is ``False``. See Also -------- CmdStanMCMC.draws CmdStanMCMC.draws_pd CmdStanGQ.draws_xr """ if not XARRAY_INSTALLED: raise RuntimeError( 'Package "xarray" is not installed, cannot produce draws array.' ) if inc_warmup and not self._save_warmup: get_logger().warning( "Draws from warmup iterations not available," ' must run sampler with "save_warmup=True".' ) if vars is None: vars_list = list(self._metadata.stan_vars.keys()) elif isinstance(vars, str): vars_list = [vars] else: vars_list = vars self._assemble_draws() num_draws = self.num_draws_sampling meta = self._metadata.cmdstan_config attrs: MutableMapping[Hashable, Any] = { "stan_version": f"{meta['stan_version_major']}." f"{meta['stan_version_minor']}.{meta['stan_version_patch']}", "model": meta["model"], "num_draws_sampling": num_draws, } if inc_warmup and self._save_warmup: num_draws += self.num_draws_warmup attrs["num_draws_warmup"] = self.num_draws_warmup data: MutableMapping[Hashable, Any] = {} coordinates: MutableMapping[Hashable, Any] = { "chain": self.chain_ids, "draw": np.arange(num_draws), } for var in vars_list: build_xarray_data( data, self._metadata.stan_vars[var], self.draws(inc_warmup=inc_warmup), ) return xr.Dataset(data, coords=coordinates, attrs=attrs).transpose( 'chain', 'draw', ... ) def stan_variable( self, var: str, inc_warmup: bool = False, ) -> np.ndarray: """ Return a numpy.ndarray which contains the set of draws for the named Stan program variable. Flattens the chains, leaving the draws in chain order. The first array dimension, corresponds to number of draws or post-warmup draws in the sample, per argument ``inc_warmup``. The remaining dimensions correspond to the shape of the Stan program variable. Underlyingly draws are in chain order, i.e., for a sample with N chains of M draws each, the first M array elements are from chain 1, the next M are from chain 2, and the last M elements are from chain N. * If the variable is a scalar variable, the return array has shape ( draws * chains, 1). * If the variable is a vector, the return array has shape ( draws * chains, len(vector)) * If the variable is a matrix, the return array has shape ( draws * chains, size(dim 1), size(dim 2) ) * If the variable is an array with N dimensions, the return array has shape ( draws * chains, size(dim 1), ..., size(dim N)) For example, if the Stan program variable ``theta`` is a 3x3 matrix, and the sample consists of 4 chains with 1000 post-warmup draws, this function will return a numpy.ndarray with shape (4000,3,3). This functionaltiy is also available via a shortcut using ``.`` - writing ``fit.a`` is a synonym for ``fit.stan_variable("a")`` :param var: variable name :param inc_warmup: When ``True`` and the warmup draws are present in the output, i.e., the sampler was run with ``save_warmup=True``, then the warmup draws are included. Default value is ``False``. See Also -------- CmdStanMCMC.stan_variables CmdStanMLE.stan_variable CmdStanPathfinder.stan_variable CmdStanVB.stan_variable CmdStanGQ.stan_variable CmdStanLaplace.stan_variable """ try: draws = self.draws(inc_warmup=inc_warmup, concat_chains=True) out: np.ndarray = self._metadata.stan_vars[var].extract_reshape( draws ) return out except KeyError: # pylint: disable=raise-missing-from raise ValueError( f'Unknown variable name: {var}\n' 'Available variables are ' + ", ".join(self._metadata.stan_vars.keys()) ) def stan_variables(self) -> Dict[str, np.ndarray]: """ Return a dictionary mapping Stan program variables names to the corresponding numpy.ndarray containing the inferred values. See Also -------- CmdStanMCMC.stan_variable CmdStanMLE.stan_variables CmdStanPathfinder.stan_variables CmdStanVB.stan_variables CmdStanGQ.stan_variables CmdStanLaplace.stan_variables """ result = {} for name in self._metadata.stan_vars: result[name] = self.stan_variable(name) return result def method_variables(self) -> Dict[str, np.ndarray]: """ Returns a dictionary of all sampler variables, i.e., all output column names ending in `__`. Assumes that all variables are scalar variables where column name is variable name. Maps each column name to a numpy.ndarray (draws x chains x 1) containing per-draw diagnostic values. """ self._assemble_draws() return { name: var.extract_reshape(self._draws) for name, var in self._metadata.method_vars.items() } def save_csvfiles(self, dir: Optional[str] = None) -> None: """ Move output CSV files to specified directory. If files were written to the temporary session directory, clean filename. E.g., save 'bernoulli-201912081451-1-5nm6as7u.csv' as 'bernoulli-201912081451-1.csv'. :param dir: directory path See Also -------- stanfit.RunSet.save_csvfiles cmdstanpy.from_csv """ self.runset.save_csvfiles(dir)