class WPDist(WESTParallelTool): prog = 'w_pdist' description = '''\ Calculate time-resolved, multi-dimensional probability distributions of WE datasets. ----------------------------------------------------------------------------- Source data ----------------------------------------------------------------------------- Source data is provided either by a user-specified function (--construct-dataset) or a list of "data set specifications" (--dsspecs). If neither is provided, the progress coordinate dataset ''pcoord'' is used. To use a custom function to extract or calculate data whose probability distribution will be calculated, specify the function in standard Python MODULE.FUNCTION syntax as the argument to --construct-dataset. This function will be called as function(n_iter,iter_group), where n_iter is the iteration whose data are being considered and iter_group is the corresponding group in the main WEST HDF5 file (west.h5). The function must return data which can be indexed as [segment][timepoint][dimension]. To use a list of data set specifications, specify --dsspecs and then list the desired datasets one-by-one (space-separated in most shells). These data set specifications are formatted as NAME[,file=FILENAME,slice=SLICE], which will use the dataset called NAME in the HDF5 file FILENAME (defaulting to the main WEST HDF5 file west.h5), and slice it with the Python slice expression SLICE (as in [0:2] to select the first two elements of the first axis of the dataset). The ``slice`` option is most useful for selecting one column (or more) from a multi-column dataset, such as arises when using a progress coordinate of multiple dimensions. ----------------------------------------------------------------------------- Histogram binning ----------------------------------------------------------------------------- By default, histograms are constructed with 100 bins in each dimension. This can be overridden by specifying -b/--bins, which accepts a number of different kinds of arguments: a single integer N N uniformly spaced bins will be used in each dimension. a sequence of integers N1,N2,... (comma-separated) N1 uniformly spaced bins will be used for the first dimension, N2 for the second, and so on. a list of lists [[B11, B12, B13, ...], [B21, B22, B23, ...], ...] The bin boundaries B11, B12, B13, ... will be used for the first dimension, B21, B22, B23, ... for the second dimension, and so on. These bin boundaries need not be uniformly spaced. These expressions will be evaluated with Python's ``eval`` construct, with ``numpy`` available for use [e.g. to specify bins using numpy.arange()]. The first two forms (integer, list of integers) will trigger a scan of all data in each dimension in order to determine the minimum and maximum values, which may be very expensive for large datasets. This can be avoided by explicitly providing bin boundaries using the list-of-lists form. Note that these bins are *NOT* at all related to the bins used to drive WE sampling. ----------------------------------------------------------------------------- Output format ----------------------------------------------------------------------------- The output file produced (specified by -o/--output, defaulting to "pdist.h5") may be fed to plothist to generate plots (or appropriately processed text or HDF5 files) from this data. In short, the following datasets are created: ``histograms`` Normalized histograms. The first axis corresponds to iteration, and remaining axes correspond to dimensions of the input dataset. ``/binbounds_0`` Vector of bin boundaries for the first (index 0) dimension. Additional datasets similarly named (/binbounds_1, /binbounds_2, ...) are created for additional dimensions. ``/midpoints_0`` Vector of bin midpoints for the first (index 0) dimension. Additional datasets similarly named are created for additional dimensions. ``n_iter`` Vector of iteration numbers corresponding to the stored histograms (i.e. the first axis of the ``histograms`` dataset). ----------------------------------------------------------------------------- Subsequent processing ----------------------------------------------------------------------------- The output generated by this program (-o/--output, default "pdist.h5") may be plotted by the ``plothist`` program. See ``plothist --help`` for more information. ----------------------------------------------------------------------------- Parallelization ----------------------------------------------------------------------------- This tool supports parallelized binning, including reading of input data. Parallel processing is the default. For simple cases (reading pre-computed input data, modest numbers of segments), serial processing (--serial) may be more efficient. ----------------------------------------------------------------------------- Command-line options ----------------------------------------------------------------------------- ''' def __init__(self): super(WPDist, self).__init__() # Parallel processing by default (this is not actually necessary, but it is # informative!) self.wm_env.default_work_manager = self.wm_env.default_parallel_work_manager # These are used throughout self.progress = ProgressIndicatorComponent() self.data_reader = WESTDataReader() self.input_dssynth = WESTDSSynthesizer(default_dsname='pcoord') self.iter_range = IterRangeSelection(self.data_reader) self.iter_range.include_args['iter_step'] = False self.binspec = None self.output_filename = None self.output_file = None self.dsspec = None self.wt_dsspec = None # dsspec for weights # These are used during histogram generation only self.iter_start = None self.iter_stop = None self.ndim = None self.ntimepoints = None self.dset_dtype = None self.binbounds = None # bin boundaries for each dimension self.midpoints = None # bin midpoints for each dimension self.data_range = None # data range for each dimension, as the pairs (min,max) self.ignore_out_of_range = False self.compress_output = False def add_args(self, parser): self.data_reader.add_args(parser) self.iter_range.add_args(parser) parser.add_argument( '-b', '--bins', dest='bins', metavar='BINEXPR', default='100', help= '''Use BINEXPR for bins. This may be an integer, which will be used for each dimension of the progress coordinate; a list of integers (formatted as [n1,n2,...]) which will use n1 bins for the first dimension, n2 for the second dimension, and so on; or a list of lists of boundaries (formatted as [[a1, a2, ...], [b1, b2, ...], ... ]), which will use [a1, a2, ...] as bin boundaries for the first dimension, [b1, b2, ...] as bin boundaries for the second dimension, and so on. (Default: 100 bins in each dimension.)''' ) parser.add_argument( '-o', '--output', dest='output', default='pdist.h5', help='''Store results in OUTPUT (default: %(default)s).''') parser.add_argument( '-C', '--compress', action='store_true', help= '''Compress histograms. May make storage of higher-dimensional histograms more tractable, at the (possible extreme) expense of increased analysis time. (Default: no compression.)''') parser.add_argument( '--loose', dest='ignore_out_of_range', action='store_true', help= '''Ignore values that do not fall within bins. (Risky, as this can make buggy bin boundaries appear as reasonable data. Only use if you are sure of your bin boundary specification.)''') igroup = parser.add_argument_group( 'input dataset options').add_mutually_exclusive_group( required=False) igroup.add_argument( '--construct-dataset', help= '''Use the given function (as in module.function) to extract source data. This function will be called once per iteration as function(n_iter, iter_group) to construct data for one iteration. Data returned must be indexable as [seg_id][timepoint][dimension]''') igroup.add_argument( '--dsspecs', nargs='+', metavar='DSSPEC', help= '''Construct probability distribution from one or more DSSPECs.''') self.progress.add_args(parser) def process_args(self, args): self.progress.process_args(args) self.data_reader.process_args(args) self.input_dssynth.h5filename = self.data_reader.we_h5filename self.input_dssynth.process_args(args) self.dsspec = self.input_dssynth.dsspec # Carrying an open HDF5 file across a fork() seems to corrupt the entire HDF5 library # Open the WEST HDF5 file just long enough to process our iteration range, then close # and reopen in go() [which executes after the fork] with self.data_reader: self.iter_range.process_args(args) self.wt_dsspec = SingleIterDSSpec(self.data_reader.we_h5filename, 'seg_index', slice=numpy.index_exp['weight']) self.binspec = args.bins self.output_filename = args.output self.ignore_out_of_range = bool(args.ignore_out_of_range) self.compress_output = args.compress or False def go(self): self.data_reader.open('r') pi = self.progress.indicator pi.operation = 'Initializing' with pi: self.output_file = h5py.File(self.output_filename, 'w') h5io.stamp_creator_data(self.output_file) self.iter_start = self.iter_range.iter_start self.iter_stop = self.iter_range.iter_stop # Construct bin boundaries self.construct_bins(self.parse_binspec(self.binspec)) for idim, (binbounds, midpoints) in enumerate( zip(self.binbounds, self.midpoints)): self.output_file['binbounds_{}'.format(idim)] = binbounds self.output_file['midpoints_{}'.format(idim)] = midpoints # construct histogram self.construct_histogram() # Record iteration range iter_range = self.iter_range.iter_range() self.output_file['n_iter'] = iter_range self.iter_range.record_data_iter_range( self.output_file['histograms']) self.output_file.close() @staticmethod def parse_binspec(binspec): namespace = {'numpy': numpy, 'inf': float('inf')} try: binspec_compiled = eval(binspec, namespace) except Exception as e: raise ValueError('invalid bin specification: {!r}'.format(e)) else: if log.isEnabledFor(logging.DEBUG): log.debug('bin specs: {!r}'.format(binspec_compiled)) return binspec_compiled def construct_bins(self, bins): ''' Construct bins according to ``bins``, which may be: 1) A scalar integer (for that number of bins in each dimension) 2) A sequence of integers (specifying number of bins for each dimension) 3) A sequence of sequences of bin boundaries (specifying boundaries for each dimension) Sets ``self.binbounds`` to a list of arrays of bin boundaries appropriate for passing to fasthist.histnd, along with ``self.midpoints`` to the midpoints of the bins. ''' if not isiterable(bins): self._construct_bins_from_scalar(bins) elif not isiterable(bins[0]): self._construct_bins_from_int_seq(bins) else: self._construct_bins_from_bound_seqs(bins) if log.isEnabledFor(logging.DEBUG): log.debug('binbounds: {!r}'.format(self.binbounds)) def scan_data_shape(self): if self.ndim is None: dset = self.dsspec.get_iter_data(self.iter_start) self.ntimepoints = dset.shape[1] self.ndim = dset.shape[2] self.dset_dtype = dset.dtype def scan_data_range(self): '''Scan input data for range in each dimension. The number of dimensions is determined from the shape of the progress coordinate as of self.iter_start.''' self.progress.indicator.new_operation('Scanning for data range', self.iter_stop - self.iter_start) self.scan_data_shape() dset_dtype = self.dset_dtype ndim = self.ndim dsspec = self.dsspec try: minval = numpy.finfo(dset_dtype).min maxval = numpy.finfo(dset_dtype).max except ValueError: minval = numpy.iinfo(dset_dtype).min maxval = numpy.iinfo(dset_dtype).max data_range = self.data_range = [(maxval, minval) for _i in range(self.ndim)] #futures = [] #for n_iter in xrange(self.iter_start, self.iter_stop): #_remote_min_max(ndim, dset_dtype, n_iter, dsspec) # futures.append(self.work_manager.submit(_remote_min_max, args=(ndim, dset_dtype, n_iter, dsspec))) #for future in self.work_manager.as_completed(futures): for future in self.work_manager.submit_as_completed( ((_remote_min_max, (ndim, dset_dtype, n_iter, dsspec), {}) for n_iter in range(self.iter_start, self.iter_stop)), self.max_queue_len): bounds = future.get_result(discard=True) for idim in range(ndim): current_min, current_max = data_range[idim] current_min = min(current_min, bounds[idim][0]) current_max = max(current_max, bounds[idim][1]) data_range[idim] = (current_min, current_max) self.progress.indicator.progress += 1 def _construct_bins_from_scalar(self, bins): if self.data_range is None: self.scan_data_range() self.binbounds = [] self.midpoints = [] for idim in range(self.ndim): lb, ub = self.data_range[idim] # Advance just beyond the upper bound of the range, so that we catch # the maximum in the histogram ub *= 1.01 boundset = numpy.linspace(lb, ub, bins + 1) midpoints = (boundset[:-1] + boundset[1:]) / 2.0 self.binbounds.append(boundset) self.midpoints.append(midpoints) def _construct_bins_from_int_seq(self, bins): if self.data_range is None: self.scan_data_range() self.binbounds = [] self.midpoints = [] for idim in range(self.ndim): lb, ub = self.data_range[idim] # Advance just beyond the upper bound of the range, so that we catch # the maximum in the histogram ub *= 1.01 boundset = numpy.linspace(lb, ub, bins[idim] + 1) midpoints = (boundset[:-1] + boundset[1:]) / 2.0 self.binbounds.append(boundset) self.midpoints.append(midpoints) def _construct_bins_from_bound_seqs(self, bins): self.binbounds = [] self.midpoints = [] for boundset in bins: boundset = numpy.asarray(boundset) if (numpy.diff(boundset) <= 0).any(): raise ValueError( 'boundary set {!r} is not strictly monotonically increasing' .format(boundset)) self.binbounds.append(boundset) self.midpoints.append((boundset[:-1] + boundset[1:]) / 2.0) def construct_histogram(self): '''Construct a histogram using bins previously constructed with ``construct_bins()``. The time series of histogram values is stored in ``histograms``. Each histogram in the time series is normalized.''' self.scan_data_shape() iter_count = self.iter_stop - self.iter_start histograms_ds = self.output_file.create_dataset( 'histograms', dtype=numpy.float64, shape=((iter_count, ) + tuple(len(bounds) - 1 for bounds in self.binbounds)), compression=9 if self.compress_output else None) binbounds = [ numpy.require(boundset, self.dset_dtype, 'C') for boundset in self.binbounds ] self.progress.indicator.new_operation('Constructing histograms', self.iter_stop - self.iter_start) task_gen = ( (_remote_bin_iter, (iiter, n_iter, self.dsspec, self.wt_dsspec, 1 if iiter > 0 else 0, binbounds, self.ignore_out_of_range), {}) for (iiter, n_iter) in enumerate(range(self.iter_start, self.iter_stop))) #futures = set() #for iiter, n_iter in enumerate(xrange(self.iter_start, self.iter_stop)): # initpoint = 1 if iiter > 0 else 0 # futures.add(self.work_manager.submit(_remote_bin_iter, # args=(iiter, n_iter, self.dsspec, self.wt_dsspec, initpoint, binbounds))) #for future in self.work_manager.as_completed(futures): #future = self.work_manager.wait_any(futures) #for future in self.work_manager.submit_as_completed(task_gen, self.queue_size): log.debug('max queue length: {!r}'.format(self.max_queue_len)) for future in self.work_manager.submit_as_completed( task_gen, self.max_queue_len): iiter, n_iter, iter_hist = future.get_result(discard=True) self.progress.indicator.progress += 1 # store histogram histograms_ds[iiter] = iter_hist del iter_hist, future
class WPDist(WESTParallelTool): prog='w_pdist' description = '''\ Calculate time-resolved, multi-dimensional probability distributions of WE datasets. ----------------------------------------------------------------------------- Source data ----------------------------------------------------------------------------- Source data is provided either by a user-specified function (--construct-dataset) or a list of "data set specifications" (--dsspecs). If neither is provided, the progress coordinate dataset ''pcoord'' is used. To use a custom function to extract or calculate data whose probability distribution will be calculated, specify the function in standard Python MODULE.FUNCTION syntax as the argument to --construct-dataset. This function will be called as function(n_iter,iter_group), where n_iter is the iteration whose data are being considered and iter_group is the corresponding group in the main WEST HDF5 file (west.h5). The function must return data which can be indexed as [segment][timepoint][dimension]. To use a list of data set specifications, specify --dsspecs and then list the desired datasets one-by-one (space-separated in most shells). These data set specifications are formatted as NAME[,file=FILENAME,slice=SLICE], which will use the dataset called NAME in the HDF5 file FILENAME (defaulting to the main WEST HDF5 file west.h5), and slice it with the Python slice expression SLICE (as in [0:2] to select the first two elements of the first axis of the dataset). The ``slice`` option is most useful for selecting one column (or more) from a multi-column dataset, such as arises when using a progress coordinate of multiple dimensions. ----------------------------------------------------------------------------- Histogram binning ----------------------------------------------------------------------------- By default, histograms are constructed with 100 bins in each dimension. This can be overridden by specifying -b/--bins, which accepts a number of different kinds of arguments: a single integer N N uniformly spaced bins will be used in each dimension. a sequence of integers N1,N2,... (comma-separated) N1 uniformly spaced bins will be used for the first dimension, N2 for the second, and so on. a list of lists [[B11, B12, B13, ...], [B21, B22, B23, ...], ...] The bin boundaries B11, B12, B13, ... will be used for the first dimension, B21, B22, B23, ... for the second dimension, and so on. These bin boundaries need not be uniformly spaced. These expressions will be evaluated with Python's ``eval`` construct, with ``numpy`` available for use [e.g. to specify bins using numpy.arange()]. The first two forms (integer, list of integers) will trigger a scan of all data in each dimension in order to determine the minimum and maximum values, which may be very expensive for large datasets. This can be avoided by explicitly providing bin boundaries using the list-of-lists form. Note that these bins are *NOT* at all related to the bins used to drive WE sampling. ----------------------------------------------------------------------------- Output format ----------------------------------------------------------------------------- The output file produced (specified by -o/--output, defaulting to "pdist.h5") may be fed to plothist to generate plots (or appropriately processed text or HDF5 files) from this data. In short, the following datasets are created: ``histograms`` Normalized histograms. The first axis corresponds to iteration, and remaining axes correspond to dimensions of the input dataset. ``/binbounds_0`` Vector of bin boundaries for the first (index 0) dimension. Additional datasets similarly named (/binbounds_1, /binbounds_2, ...) are created for additional dimensions. ``/midpoints_0`` Vector of bin midpoints for the first (index 0) dimension. Additional datasets similarly named are created for additional dimensions. ``n_iter`` Vector of iteration numbers corresponding to the stored histograms (i.e. the first axis of the ``histograms`` dataset). ----------------------------------------------------------------------------- Subsequent processing ----------------------------------------------------------------------------- The output generated by this program (-o/--output, default "pdist.h5") may be plotted by the ``plothist`` program. See ``plothist --help`` for more information. ----------------------------------------------------------------------------- Parallelization ----------------------------------------------------------------------------- This tool supports parallelized binning, including reading of input data. Parallel processing is the default. For simple cases (reading pre-computed input data, modest numbers of segments), serial processing (--serial) may be more efficient. ----------------------------------------------------------------------------- Command-line options ----------------------------------------------------------------------------- ''' def __init__(self): super(WPDist,self).__init__() # Parallel processing by default (this is not actually necessary, but it is # informative!) self.wm_env.default_work_manager = self.wm_env.default_parallel_work_manager # These are used throughout self.progress = ProgressIndicatorComponent() self.data_reader = WESTDataReader() self.input_dssynth = WESTDSSynthesizer(default_dsname='pcoord') self.iter_range = IterRangeSelection(self.data_reader) self.iter_range.include_args['iter_step'] = False self.binspec = None self.output_filename = None self.output_file = None self.dsspec = None self.wt_dsspec = None # dsspec for weights # These are used during histogram generation only self.iter_start = None self.iter_stop = None self.ndim = None self.ntimepoints = None self.dset_dtype = None self.binbounds = None # bin boundaries for each dimension self.midpoints = None # bin midpoints for each dimension self.data_range = None # data range for each dimension, as the pairs (min,max) self.ignore_out_of_range = False self.compress_output = False def add_args(self, parser): self.data_reader.add_args(parser) self.iter_range.add_args(parser) parser.add_argument('-b', '--bins', dest='bins', metavar='BINEXPR', default='100', help='''Use BINEXPR for bins. This may be an integer, which will be used for each dimension of the progress coordinate; a list of integers (formatted as [n1,n2,...]) which will use n1 bins for the first dimension, n2 for the second dimension, and so on; or a list of lists of boundaries (formatted as [[a1, a2, ...], [b1, b2, ...], ... ]), which will use [a1, a2, ...] as bin boundaries for the first dimension, [b1, b2, ...] as bin boundaries for the second dimension, and so on. (Default: 100 bins in each dimension.)''') parser.add_argument('-o', '--output', dest='output', default='pdist.h5', help='''Store results in OUTPUT (default: %(default)s).''') parser.add_argument('-C', '--compress', action='store_true', help='''Compress histograms. May make storage of higher-dimensional histograms more tractable, at the (possible extreme) expense of increased analysis time. (Default: no compression.)''') parser.add_argument('--loose', dest='ignore_out_of_range', action='store_true', help='''Ignore values that do not fall within bins. (Risky, as this can make buggy bin boundaries appear as reasonable data. Only use if you are sure of your bin boundary specification.)''') igroup = parser.add_argument_group('input dataset options').add_mutually_exclusive_group(required=False) igroup.add_argument('--construct-dataset', help='''Use the given function (as in module.function) to extract source data. This function will be called once per iteration as function(n_iter, iter_group) to construct data for one iteration. Data returned must be indexable as [seg_id][timepoint][dimension]''') igroup.add_argument('--dsspecs', nargs='+', metavar='DSSPEC', help='''Construct probability distribution from one or more DSSPECs.''') self.progress.add_args(parser) def process_args(self, args): self.progress.process_args(args) self.data_reader.process_args(args) self.input_dssynth.h5filename = self.data_reader.we_h5filename self.input_dssynth.process_args(args) self.dsspec = self.input_dssynth.dsspec # Carrying an open HDF5 file across a fork() seems to corrupt the entire HDF5 library # Open the WEST HDF5 file just long enough to process our iteration range, then close # and reopen in go() [which executes after the fork] with self.data_reader: self.iter_range.process_args(args) self.wt_dsspec = SingleIterDSSpec(self.data_reader.we_h5filename, 'seg_index', slice=numpy.index_exp['weight']) self.binspec = args.bins self.output_filename = args.output self.ignore_out_of_range = bool(args.ignore_out_of_range) self.compress_output = args.compress or False def go(self): self.data_reader.open('r') pi = self.progress.indicator pi.operation = 'Initializing' with pi: self.output_file = h5py.File(self.output_filename, 'w') h5io.stamp_creator_data(self.output_file) self.iter_start = self.iter_range.iter_start self.iter_stop = self.iter_range.iter_stop # Construct bin boundaries self.construct_bins(self.parse_binspec(self.binspec)) for idim, (binbounds, midpoints) in enumerate(izip(self.binbounds, self.midpoints)): self.output_file['binbounds_{}'.format(idim)] = binbounds self.output_file['midpoints_{}'.format(idim)] = midpoints # construct histogram self.construct_histogram() # Record iteration range iter_range = self.iter_range.iter_range() self.output_file['n_iter'] = iter_range self.iter_range.record_data_iter_range(self.output_file['histograms']) self.output_file.close() @staticmethod def parse_binspec(binspec): namespace = {'numpy': numpy, 'inf': float('inf')} try: binspec_compiled = eval(binspec,namespace) except Exception as e: raise ValueError('invalid bin specification: {!r}'.format(e)) else: if log.isEnabledFor(logging.DEBUG): log.debug('bin specs: {!r}'.format(binspec_compiled)) return binspec_compiled def construct_bins(self, bins): ''' Construct bins according to ``bins``, which may be: 1) A scalar integer (for that number of bins in each dimension) 2) A sequence of integers (specifying number of bins for each dimension) 3) A sequence of sequences of bin boundaries (specifying boundaries for each dimension) Sets ``self.binbounds`` to a list of arrays of bin boundaries appropriate for passing to fasthist.histnd, along with ``self.midpoints`` to the midpoints of the bins. ''' if not isiterable(bins): self._construct_bins_from_scalar(bins) elif not isiterable(bins[0]): self._construct_bins_from_int_seq(bins) else: self._construct_bins_from_bound_seqs(bins) if log.isEnabledFor(logging.DEBUG): log.debug('binbounds: {!r}'.format(self.binbounds)) def scan_data_shape(self): if self.ndim is None: dset = self.dsspec.get_iter_data(self.iter_start) self.ntimepoints = dset.shape[1] self.ndim = dset.shape[2] self.dset_dtype = dset.dtype def scan_data_range(self): '''Scan input data for range in each dimension. The number of dimensions is determined from the shape of the progress coordinate as of self.iter_start.''' self.progress.indicator.new_operation('Scanning for data range', self.iter_stop-self.iter_start) self.scan_data_shape() dset_dtype = self.dset_dtype ndim = self.ndim dsspec = self.dsspec try: minval = numpy.finfo(dset_dtype).min maxval = numpy.finfo(dset_dtype).max except ValueError: minval = numpy.iinfo(dset_dtype).min maxval = numpy.iinfo(dset_dtype).max data_range = self.data_range = [(maxval,minval) for _i in xrange(self.ndim)] #futures = [] #for n_iter in xrange(self.iter_start, self.iter_stop): #_remote_min_max(ndim, dset_dtype, n_iter, dsspec) # futures.append(self.work_manager.submit(_remote_min_max, args=(ndim, dset_dtype, n_iter, dsspec))) #for future in self.work_manager.as_completed(futures): for future in self.work_manager.submit_as_completed(((_remote_min_max, (ndim, dset_dtype, n_iter, dsspec), {}) for n_iter in xrange(self.iter_start, self.iter_stop)), self.max_queue_len): bounds = future.get_result(discard=True) for idim in xrange(ndim): current_min, current_max = data_range[idim] current_min = min(current_min, bounds[idim][0]) current_max = max(current_max, bounds[idim][1]) data_range[idim] = (current_min, current_max) self.progress.indicator.progress += 1 def _construct_bins_from_scalar(self, bins): if self.data_range is None: self.scan_data_range() self.binbounds = [] self.midpoints = [] for idim in xrange(self.ndim): lb, ub = self.data_range[idim] # Advance just beyond the upper bound of the range, so that we catch # the maximum in the histogram ub *= 1.01 boundset = numpy.linspace(lb,ub,bins+1) midpoints = (boundset[:-1] + boundset[1:]) / 2.0 self.binbounds.append(boundset) self.midpoints.append(midpoints) def _construct_bins_from_int_seq(self, bins): if self.data_range is None: self.scan_data_range() self.binbounds = [] self.midpoints = [] for idim in xrange(self.ndim): lb, ub = self.data_range[idim] # Advance just beyond the upper bound of the range, so that we catch # the maximum in the histogram ub *= 1.01 boundset = numpy.linspace(lb,ub,bins[idim]+1) midpoints = (boundset[:-1] + boundset[1:]) / 2.0 self.binbounds.append(boundset) self.midpoints.append(midpoints) def _construct_bins_from_bound_seqs(self, bins): self.binbounds = [] self.midpoints = [] for boundset in bins: boundset = numpy.asarray(boundset) if (numpy.diff(boundset) <= 0).any(): raise ValueError('boundary set {!r} is not strictly monotonically increasing'.format(boundset)) self.binbounds.append(boundset) self.midpoints.append((boundset[:-1]+boundset[1:])/2.0) def construct_histogram(self): '''Construct a histogram using bins previously constructed with ``construct_bins()``. The time series of histogram values is stored in ``histograms``. Each histogram in the time series is normalized.''' self.scan_data_shape() iter_count = self.iter_stop - self.iter_start histograms_ds = self.output_file.create_dataset('histograms', dtype=numpy.float64, shape=((iter_count,) + tuple(len(bounds)-1 for bounds in self.binbounds)), compression=9 if self.compress_output else None) binbounds = [numpy.require(boundset, self.dset_dtype, 'C') for boundset in self.binbounds] self.progress.indicator.new_operation('Constructing histograms',self.iter_stop-self.iter_start) task_gen = ((_remote_bin_iter, (iiter, n_iter, self.dsspec, self.wt_dsspec, 1 if iiter > 0 else 0, binbounds, self.ignore_out_of_range), {}) for (iiter,n_iter) in enumerate(xrange(self.iter_start, self.iter_stop))) #futures = set() #for iiter, n_iter in enumerate(xrange(self.iter_start, self.iter_stop)): # initpoint = 1 if iiter > 0 else 0 # futures.add(self.work_manager.submit(_remote_bin_iter, # args=(iiter, n_iter, self.dsspec, self.wt_dsspec, initpoint, binbounds))) #for future in self.work_manager.as_completed(futures): #future = self.work_manager.wait_any(futures) #for future in self.work_manager.submit_as_completed(task_gen, self.queue_size): log.debug('max queue length: {!r}'.format(self.max_queue_len)) for future in self.work_manager.submit_as_completed(task_gen, self.max_queue_len): iiter, n_iter, iter_hist = future.get_result(discard=True) self.progress.indicator.progress += 1 # store histogram histograms_ds[iiter] = iter_hist del iter_hist, future
class WFluxanlTool(WESTTool): prog = 'w_fluxanl' description = '''\ Extract fluxes into pre-defined target states from WEST data, average, and construct confidence intervals. Monte Carlo bootstrapping is used to account for the correlated and possibly non-Gaussian statistical error in flux measurements. All non-graphical output (including that to the terminal and HDF5) assumes that the propagation/resampling period ``tau`` is equal to unity; to obtain results in familiar units, divide all fluxes and multiply all correlation lengths by the true value of ``tau``. ''' output_format_version = 2 def __init__(self): super(WFluxanlTool, self).__init__() self.data_reader = WESTDataReader() self.iter_range = IterRangeSelection() self.output_h5file = None self.output_group = None self.target_groups = {} self.fluxdata = {} self.alpha = None self.autocorrel_alpha = None self.n_sets = None self.do_evol = False self.evol_step = 1 def add_args(self, parser): self.data_reader.add_args(parser) self.iter_range.add_args(parser) ogroup = parser.add_argument_group('output options') ogroup.add_argument( '-o', '--output', default='fluxanl.h5', help= 'Store intermediate data and analysis results to OUTPUT (default: %(default)s).' ) cgroup = parser.add_argument_group('calculation options') cgroup.add_argument( '--disable-bootstrap', '-db', dest='bootstrap', action='store_const', const=False, help='''Enable the use of Monte Carlo Block Bootstrapping.''') cgroup.add_argument('--disable-correl', '-dc', dest='correl', action='store_const', const=False, help='''Disable the correlation analysis.''') cgroup.add_argument( '-a', '--alpha', type=float, default=0.05, help= '''Calculate a (1-ALPHA) confidence interval on the average flux' (default: %(default)s)''') cgroup.add_argument( '--autocorrel-alpha', type=float, dest='acalpha', metavar='ACALPHA', help='''Evaluate autocorrelation of flux to (1-ACALPHA) significance. Note that too small an ACALPHA will result in failure to detect autocorrelation in a noisy flux signal. (Default: same as ALPHA.)''' ) cgroup.add_argument( '-N', '--nsets', type=int, help= '''Use NSETS samples for bootstrapping (default: chosen based on ALPHA)''' ) cgroup.add_argument( '--evol', action='store_true', dest='do_evol', help= '''Calculate time evolution of flux confidence intervals (expensive).''' ) cgroup.add_argument( '--evol-step', type=int, default=1, metavar='ESTEP', help= '''Calculate time evolution of flux confidence intervals every ESTEP iterations (default: %(default)s)''') def process_args(self, args): self.data_reader.process_args(args) self.data_reader.open() self.iter_range.data_manager = self.data_reader self.iter_range.process_args(args) self.output_h5file = h5py.File(args.output, 'w') self.alpha = args.alpha # Disable the bootstrap or the correlation analysis. self.mcbs_enable = args.bootstrap if args.bootstrap is not None else True self.do_correl = args.correl if args.correl is not None else True self.autocorrel_alpha = args.acalpha or self.alpha self.n_sets = args.nsets or mclib.get_bssize(self.alpha) self.do_evol = args.do_evol self.evol_step = args.evol_step or 1 def calc_store_flux_data(self): westpa.rc.pstatus( 'Calculating mean flux and confidence intervals for iterations [{},{})' .format(self.iter_range.iter_start, self.iter_range.iter_stop)) fluxdata = extract_fluxes(self.iter_range.iter_start, self.iter_range.iter_stop, self.data_reader) # Create a group to store data in output_group = h5io.create_hdf5_group(self.output_h5file, 'target_flux', replace=False, creating_program=self.prog) self.output_group = output_group output_group.attrs['version_code'] = self.output_format_version self.iter_range.record_data_iter_range(output_group) n_targets = len(fluxdata) index = numpy.empty((len(fluxdata), ), dtype=target_index_dtype) avg_fluxdata = numpy.empty((n_targets, ), dtype=ci_dtype) for itarget, (target_label, target_fluxdata) in enumerate(fluxdata.items()): # Create group and index entry index[itarget]['target_label'] = str(target_label) target_group = output_group.create_group( 'target_{}'.format(itarget)) self.target_groups[target_label] = target_group # Store per-iteration values target_group['n_iter'] = target_fluxdata['n_iter'] target_group['count'] = target_fluxdata['count'] target_group['flux'] = target_fluxdata['flux'] h5io.label_axes(target_group['flux'], ['n_iter'], units=['tau^-1']) # Calculate flux autocorrelation fluxes = target_fluxdata['flux'] mean_flux = fluxes.mean() fmm = fluxes - mean_flux acorr = fftconvolve(fmm, fmm[::-1]) acorr = acorr[len(acorr) // 2:] acorr /= acorr[0] acorr_ds = target_group.create_dataset('flux_autocorrel', data=acorr) h5io.label_axes(acorr_ds, ['lag'], ['tau']) # Calculate overall averages and CIs #avg, lb_ci, ub_ci, correl_len = mclib.mcbs_ci_correl(fluxes, numpy.mean, self.alpha, self.n_sets, # autocorrel_alpha=self.autocorrel_alpha, subsample=numpy.mean) avg, lb_ci, ub_ci, sterr, correl_len = mclib.mcbs_ci_correl( {'dataset': fluxes}, estimator=(lambda stride, dataset: numpy.mean(dataset)), alpha=self.alpha, n_sets=self.n_sets, autocorrel_alpha=self.autocorrel_alpha, subsample=numpy.mean, do_correl=self.do_correl, mcbs_enable=self.mcbs_enable) avg_fluxdata[itarget] = (self.iter_range.iter_start, self.iter_range.iter_stop, avg, lb_ci, ub_ci, sterr, correl_len) westpa.rc.pstatus('target {!r}:'.format(target_label)) westpa.rc.pstatus( ' correlation length = {} tau'.format(correl_len)) westpa.rc.pstatus( ' mean flux and CI = {:e} ({:e},{:e}) tau^(-1)'.format( avg, lb_ci, ub_ci)) index[itarget]['mean_flux'] = avg index[itarget]['mean_flux_ci_lb'] = lb_ci index[itarget]['mean_flux_ci_ub'] = ub_ci index[itarget]['mean_flux_correl_len'] = correl_len # Write index and summary index_ds = output_group.create_dataset('index', data=index) index_ds.attrs['mcbs_alpha'] = self.alpha index_ds.attrs['mcbs_autocorrel_alpha'] = self.autocorrel_alpha index_ds.attrs['mcbs_n_sets'] = self.n_sets self.fluxdata = fluxdata self.output_h5file['avg_flux'] = avg_fluxdata def calc_evol_flux(self): westpa.rc.pstatus( 'Calculating cumulative evolution of flux confidence intervals every {} iteration(s)' .format(self.evol_step)) for itarget, (target_label, target_fluxdata) in enumerate(self.fluxdata.items()): fluxes = target_fluxdata['flux'] target_group = self.target_groups[target_label] iter_start = target_group['n_iter'][0] iter_stop = target_group['n_iter'][-1] iter_count = iter_stop - iter_start n_blocks = iter_count // self.evol_step if iter_count % self.evol_step > 0: n_blocks += 1 cis = numpy.empty((n_blocks, ), dtype=ci_dtype) for iblock in range(n_blocks): block_iter_stop = min( iter_start + (iblock + 1) * self.evol_step, iter_stop) istop = min((iblock + 1) * self.evol_step, len(target_fluxdata['flux'])) fluxes = target_fluxdata['flux'][:istop] #avg, ci_lb, ci_ub, correl_len = mclib.mcbs_ci_correl(fluxes, numpy.mean, self.alpha, self.n_sets, # autocorrel_alpha = self.autocorrel_alpha, # subsample=numpy.mean) avg, ci_lb, ci_ub, sterr, correl_len = mclib.mcbs_ci_correl( {'dataset': fluxes}, estimator=(lambda stride, dataset: numpy.mean(dataset)), alpha=self.alpha, n_sets=self.n_sets, autocorrel_alpha=self.autocorrel_alpha, subsample=numpy.mean, do_correl=self.do_correl, mcbs_enable=self.mcbs_enable) cis[iblock]['iter_start'] = iter_start cis[iblock]['iter_stop'] = block_iter_stop cis[iblock]['expected'], cis[iblock]['ci_lbound'], cis[iblock][ 'ci_ubound'] = avg, ci_lb, ci_ub cis[iblock]['corr_len'] = correl_len cis[iblock]['sterr'] = sterr del fluxes cis_ds = target_group.create_dataset('flux_evolution', data=cis) cis_ds.attrs['iter_step'] = self.evol_step cis_ds.attrs['mcbs_alpha'] = self.alpha cis_ds.attrs['mcbs_autocorrel_alpha'] = self.autocorrel_alpha cis_ds.attrs['mcbs_n_sets'] = self.n_sets def go(self): self.calc_store_flux_data() if self.do_evol: self.calc_evol_flux()
class WFluxanlTool(WESTTool): prog='w_fluxanl' description = '''\ Extract fluxes into pre-defined target states from WEST data, average, and construct confidence intervals. Monte Carlo bootstrapping is used to account for the correlated and possibly non-Gaussian statistical error in flux measurements. All non-graphical output (including that to the terminal and HDF5) assumes that the propagation/resampling period ``tau`` is equal to unity; to obtain results in familiar units, divide all fluxes and multiply all correlation lengths by the true value of ``tau``. ''' output_format_version = 2 def __init__(self): super(WFluxanlTool,self).__init__() self.data_reader = WESTDataReader() self.iter_range = IterRangeSelection() self.output_h5file = None self.output_group = None self.target_groups = {} self.fluxdata = {} self.alpha = None self.autocorrel_alpha = None self.n_sets = None self.do_evol = False self.evol_step = 1 def add_args(self, parser): self.data_reader.add_args(parser) self.iter_range.add_args(parser) ogroup = parser.add_argument_group('output options') ogroup.add_argument('-o', '--output', default='fluxanl.h5', help='Store intermediate data and analysis results to OUTPUT (default: %(default)s).') cgroup = parser.add_argument_group('calculation options') cgroup.add_argument('--disable-bootstrap', '-db', dest='bootstrap', action='store_const', const=False, help='''Enable the use of Monte Carlo Block Bootstrapping.''') cgroup.add_argument('--disable-correl', '-dc', dest='correl', action='store_const', const=False, help='''Disable the correlation analysis.''') cgroup.add_argument('-a', '--alpha', type=float, default=0.05, help='''Calculate a (1-ALPHA) confidence interval on the average flux' (default: %(default)s)''') cgroup.add_argument('--autocorrel-alpha', type=float, dest='acalpha', metavar='ACALPHA', help='''Evaluate autocorrelation of flux to (1-ACALPHA) significance. Note that too small an ACALPHA will result in failure to detect autocorrelation in a noisy flux signal. (Default: same as ALPHA.)''') cgroup.add_argument('-N', '--nsets', type=int, help='''Use NSETS samples for bootstrapping (default: chosen based on ALPHA)''') cgroup.add_argument('--evol', action='store_true', dest='do_evol', help='''Calculate time evolution of flux confidence intervals (expensive).''') cgroup.add_argument('--evol-step', type=int, default=1, metavar='ESTEP', help='''Calculate time evolution of flux confidence intervals every ESTEP iterations (default: %(default)s)''') def process_args(self, args): self.data_reader.process_args(args) self.data_reader.open() self.iter_range.data_manager = self.data_reader self.iter_range.process_args(args) self.output_h5file = h5py.File(args.output, 'w') self.alpha = args.alpha # Disable the bootstrap or the correlation analysis. self.mcbs_enable = args.bootstrap if args.bootstrap is not None else True self.do_correl = args.correl if args.correl is not None else True self.autocorrel_alpha = args.acalpha or self.alpha self.n_sets = args.nsets or mclib.get_bssize(self.alpha) self.do_evol = args.do_evol self.evol_step = args.evol_step or 1 def calc_store_flux_data(self): westpa.rc.pstatus('Calculating mean flux and confidence intervals for iterations [{},{})' .format(self.iter_range.iter_start, self.iter_range.iter_stop)) fluxdata = extract_fluxes(self.iter_range.iter_start, self.iter_range.iter_stop, self.data_reader) # Create a group to store data in output_group = h5io.create_hdf5_group(self.output_h5file, 'target_flux', replace=False, creating_program=self.prog) self.output_group = output_group output_group.attrs['version_code'] = self.output_format_version self.iter_range.record_data_iter_range(output_group) n_targets = len(fluxdata) index = numpy.empty((len(fluxdata),), dtype=target_index_dtype) avg_fluxdata = numpy.empty((n_targets,), dtype=ci_dtype) for itarget, (target_label, target_fluxdata) in enumerate(fluxdata.iteritems()): # Create group and index entry index[itarget]['target_label'] = str(target_label) target_group = output_group.create_group('target_{}'.format(itarget)) self.target_groups[target_label] = target_group # Store per-iteration values target_group['n_iter'] = target_fluxdata['n_iter'] target_group['count'] = target_fluxdata['count'] target_group['flux'] = target_fluxdata['flux'] h5io.label_axes(target_group['flux'], ['n_iter'], units=['tau^-1']) # Calculate flux autocorrelation fluxes = target_fluxdata['flux'] mean_flux = fluxes.mean() fmm = fluxes - mean_flux acorr = fftconvolve(fmm,fmm[::-1]) acorr = acorr[len(acorr)//2:] acorr /= acorr[0] acorr_ds = target_group.create_dataset('flux_autocorrel', data=acorr) h5io.label_axes(acorr_ds, ['lag'], ['tau']) # Calculate overall averages and CIs #avg, lb_ci, ub_ci, correl_len = mclib.mcbs_ci_correl(fluxes, numpy.mean, self.alpha, self.n_sets, # autocorrel_alpha=self.autocorrel_alpha, subsample=numpy.mean) avg, lb_ci, ub_ci, sterr, correl_len = mclib.mcbs_ci_correl({'dataset': fluxes}, estimator=(lambda stride, dataset: numpy.mean(dataset)), alpha=self.alpha, n_sets=self.n_sets, autocorrel_alpha=self.autocorrel_alpha, subsample=numpy.mean, do_correl=self.do_correl, mcbs_enable=self.mcbs_enable ) avg_fluxdata[itarget] = (self.iter_range.iter_start, self.iter_range.iter_stop, avg, lb_ci, ub_ci, sterr, correl_len) westpa.rc.pstatus('target {!r}:'.format(target_label)) westpa.rc.pstatus(' correlation length = {} tau'.format(correl_len)) westpa.rc.pstatus(' mean flux and CI = {:e} ({:e},{:e}) tau^(-1)'.format(avg,lb_ci,ub_ci)) index[itarget]['mean_flux'] = avg index[itarget]['mean_flux_ci_lb'] = lb_ci index[itarget]['mean_flux_ci_ub'] = ub_ci index[itarget]['mean_flux_correl_len'] = correl_len # Write index and summary index_ds = output_group.create_dataset('index', data=index) index_ds.attrs['mcbs_alpha'] = self.alpha index_ds.attrs['mcbs_autocorrel_alpha'] = self.autocorrel_alpha index_ds.attrs['mcbs_n_sets'] = self.n_sets self.fluxdata = fluxdata self.output_h5file['avg_flux'] = avg_fluxdata def calc_evol_flux(self): westpa.rc.pstatus('Calculating cumulative evolution of flux confidence intervals every {} iteration(s)' .format(self.evol_step)) for itarget, (target_label, target_fluxdata) in enumerate(self.fluxdata.iteritems()): fluxes = target_fluxdata['flux'] target_group = self.target_groups[target_label] iter_start = target_group['n_iter'][0] iter_stop = target_group['n_iter'][-1] iter_count = iter_stop - iter_start n_blocks = iter_count // self.evol_step if iter_count % self.evol_step > 0: n_blocks += 1 cis = numpy.empty((n_blocks,), dtype=ci_dtype) for iblock in xrange(n_blocks): block_iter_stop = min(iter_start + (iblock+1)*self.evol_step, iter_stop) istop = min((iblock+1)*self.evol_step, len(target_fluxdata['flux'])) fluxes = target_fluxdata['flux'][:istop] #avg, ci_lb, ci_ub, correl_len = mclib.mcbs_ci_correl(fluxes, numpy.mean, self.alpha, self.n_sets, # autocorrel_alpha = self.autocorrel_alpha, # subsample=numpy.mean) avg, ci_lb, ci_ub, sterr, correl_len = mclib.mcbs_ci_correl({'dataset': fluxes}, estimator=(lambda stride, dataset: numpy.mean(dataset)), alpha=self.alpha, n_sets=self.n_sets, autocorrel_alpha = self.autocorrel_alpha, subsample=numpy.mean, do_correl=self.do_correl, mcbs_enable=self.mcbs_enable ) cis[iblock]['iter_start'] = iter_start cis[iblock]['iter_stop'] = block_iter_stop cis[iblock]['expected'], cis[iblock]['ci_lbound'], cis[iblock]['ci_ubound'] = avg, ci_lb, ci_ub cis[iblock]['corr_len'] = correl_len cis[iblock]['sterr'] = sterr del fluxes cis_ds = target_group.create_dataset('flux_evolution', data=cis) cis_ds.attrs['iter_step'] = self.evol_step cis_ds.attrs['mcbs_alpha'] = self.alpha cis_ds.attrs['mcbs_autocorrel_alpha'] = self.autocorrel_alpha cis_ds.attrs['mcbs_n_sets'] = self.n_sets def go(self): self.calc_store_flux_data() if self.do_evol: self.calc_evol_flux()