"""Define the Component class."""
import sys
from collections import defaultdict
from collections.abc import Iterable
from itertools import product
from numbers import Integral
import numpy as np
from numpy import ndarray, isscalar, ndim, atleast_1d, atleast_2d, promote_types
from scipy.sparse import issparse, coo_matrix
from openmdao.core.system import System, _supported_methods, _DEFAULT_COLORING_META, \
global_meta_names, collect_errors
from openmdao.core.constants import INT_DTYPE
from openmdao.jacobians.dictionary_jacobian import DictionaryJacobian
from openmdao.utils.array_utils import shape_to_len
from openmdao.utils.units import simplify_unit
from openmdao.utils.name_maps import abs_key_iter, abs_key2rel_key, rel_name2abs_name
from openmdao.utils.mpi import MPI
from openmdao.utils.general_utils import format_as_float_or_array, ensure_compatible, \
find_matches, make_set, convert_src_inds, inconsistent_across_procs
from openmdao.utils.indexer import Indexer, indexer
import openmdao.utils.coloring as coloring_mod
from openmdao.utils.om_warnings import issue_warning, MPIWarning, DistributedComponentWarning, \
DerivativesWarning, warn_deprecation
_forbidden_chars = {'.', '*', '?', '!', '[', ']'}
_whitespace = {' ', '\t', '\r', '\n'}
_allowed_types = (list, tuple, ndarray, Iterable)
def _valid_var_name(name):
"""
Determine if the proposed name is a valid variable name.
Leading and trailing whitespace is illegal, and a specific list of characters
are illegal anywhere in the string.
Parameters
----------
name : str
Proposed name.
Returns
-------
bool
True if the proposed name is a valid variable name, else False.
"""
global _forbidden_chars, _whitespace
if not name:
return False
if _forbidden_chars.intersection(name):
return False
return name is name.strip()
[docs]class Component(System):
"""
Base Component class; not to be directly instantiated.
Parameters
----------
**kwargs : dict of keyword arguments
Available here and in all descendants of this system.
Attributes
----------
_var_rel2meta : dict
Dictionary mapping relative names to metadata.
This is only needed while adding inputs and outputs. During setup, these are used to
build the dictionaries of metadata.
_static_var_rel2meta : dict
Static version of above - stores data for variables added outside of setup.
_var_rel_names : {'input': [str, ...], 'output': [str, ...]}
List of relative names of owned variables existing on current proc.
This is only needed while adding inputs and outputs. During setup, these are used to
determine the list of absolute names.
_static_var_rel_names : dict
Static version of above - stores names of variables added outside of setup.
_declared_partials : dict
Cached storage of user-declared partials.
_declared_partial_checks : list
Cached storage of user-declared check partial options.
_no_check_partials : bool
If True, the check_partials function will ignore this component.
_has_distrib_outputs : bool
If True, this component has at least one distributed output.
"""
[docs] def __init__(self, **kwargs):
"""
Initialize all attributes.
"""
super().__init__(**kwargs)
self._var_rel_names = {'input': [], 'output': []}
self._var_rel2meta = {}
self._static_var_rel_names = {'input': [], 'output': []}
self._static_var_rel2meta = {}
self._declared_partials = defaultdict(dict)
self._declared_partial_checks = []
self._no_check_partials = False
self._has_distrib_outputs = False
def _declare_options(self):
"""
Declare options before kwargs are processed in the init method.
"""
super()._declare_options()
self.options.declare('distributed', types=bool, default=False,
desc='If True, set all variables in this component as distributed '
'across multiple processes')
self.options.declare('run_root_only', types=bool, default=False,
desc='If True, call compute, compute_partials, linearize, '
'apply_linear, apply_nonlinear, and compute_jacvec_product '
'only on rank 0 and broadcast the results to the other ranks.')
def _check_matfree_deprecation(self):
# check for mixed distributed variables
has_dist_ins = has_nd_ins = has_dist_outs = has_nd_outs = False
for name in self._var_rel_names['input']:
meta = self._var_rel2meta[name]
if meta['distributed']:
has_dist_ins = True
else:
has_nd_ins = True
for name in self._var_rel_names['output']:
meta = self._var_rel2meta[name]
if meta['distributed']:
has_dist_outs = True
else:
has_nd_outs = True
if (has_nd_ins and has_dist_outs) or (has_dist_ins and has_nd_outs):
warn_deprecation(f"{self.msginfo}: It appears this component mixes "
"distributed/non-distributed inputs and outputs, so it may break "
"starting with OpenMDAO 3.25, where the convention "
"used when passing data between distributed and non-distributed "
"inputs and outputs within a matrix free component will change. "
"See https://github.com/OpenMDAO/POEMs/blob/master/POEM_075.md for "
"details.")
[docs] def setup(self):
"""
Declare inputs and outputs.
Available attributes:
name
pathname
comm
options
"""
pass
def _setup_procs(self, pathname, comm, mode, prob_meta):
"""
Execute first phase of the setup process.
Distribute processors, assign pathnames, and call setup on the component.
Parameters
----------
pathname : str
Global name of the system, including the path.
comm : MPI.Comm or <FakeComm>
MPI communicator object.
mode : str
Derivatives calculation mode, 'fwd' for forward, and 'rev' for
reverse (adjoint). Default is 'rev'.
prob_meta : dict
Problem level metadata.
"""
super()._setup_procs(pathname, comm, mode, prob_meta)
if self._num_par_fd > 1:
if comm.size > 1:
comm = self._setup_par_fd_procs(comm)
elif not MPI:
issue_warning(f"MPI is not active but num_par_fd = {self._num_par_fd}. No parallel "
"finite difference will be performed.",
prefix=self.msginfo, category=MPIWarning)
self._num_par_fd = 1
self.comm = comm
nprocs = comm.size
# Clear out old variable information so that we can call setup on the component.
self._var_rel_names = {'input': [], 'output': []}
self._var_rel2meta = {}
if comm.size == 1:
self._has_distrib_vars = self._has_distrib_outputs = False
for meta in self._static_var_rel2meta.values():
# variable isn't distributed if we're only running on 1 proc
if nprocs == 1 and 'distributed' in meta and meta['distributed']:
meta['distributed'] = False
# reset shape if any dynamic shape parameters are set in case this is a resetup
# NOTE: this is necessary because we allow variables to be added in __init__.
if 'shape_by_conn' in meta and (meta['shape_by_conn'] or
meta['copy_shape'] is not None):
meta['shape'] = None
if not isscalar(meta['val']):
if meta['val'].size > 0:
meta['val'] = meta['val'].flatten()[0]
else:
meta['val'] = 1.0
self._var_rel2meta.update(self._static_var_rel2meta)
for io in ['input', 'output']:
self._var_rel_names[io].extend(self._static_var_rel_names[io])
self.setup()
self._setup_check()
self._set_vector_class()
def _set_vector_class(self):
if self._has_distrib_vars:
dist_vec_class = self._problem_meta['distributed_vector_class']
if dist_vec_class is not None:
self._vector_class = dist_vec_class
else:
issue_warning("Component contains distributed variables, "
"but there is no distributed vector implementation (MPI/PETSc) "
"available. The default non-distributed vectors will be used.",
prefix=self.msginfo, category=DistributedComponentWarning)
self._vector_class = self._problem_meta['local_vector_class']
else:
self._vector_class = self._problem_meta['local_vector_class']
def _configure_check(self):
"""
Do any error checking on i/o configuration.
"""
# Check here if declare_coloring was called during setup but declare_partials wasn't.
# If declare partials wasn't called, call it with of='*' and wrt='*' so we'll have
# something to color.
if self._coloring_info['coloring'] is not None:
for key, meta in self._declared_partials.items():
if 'method' in meta and meta['method'] is not None:
break
else:
method = self._coloring_info['method']
issue_warning("declare_coloring or use_fixed_coloring was called but no approx"
" partials were declared. Declaring all partials as approximated "
f"using default metadata and method='{method}'.", prefix=self.msginfo,
category=DerivativesWarning)
self.declare_partials('*', '*', method=method)
super()._configure_check()
def _setup_var_data(self):
"""
Compute the list of abs var names, abs/prom name maps, and metadata dictionaries.
"""
global global_meta_names
super()._setup_var_data()
allprocs_prom2abs_list = self._var_allprocs_prom2abs_list
abs2prom = self._var_allprocs_abs2prom = self._var_abs2prom
abs_in2prom_info = self._problem_meta['abs_in2prom_info']
# Compute the prefix for turning rel/prom names into abs names
prefix = self.pathname + '.' if self.pathname else ''
for io in ['input', 'output']:
abs2meta = self._var_abs2meta[io]
allprocs_abs2meta = self._var_allprocs_abs2meta[io]
is_input = io == 'input'
for prom_name in self._var_rel_names[io]:
abs_name = prefix + prom_name
abs2meta[abs_name] = metadata = self._var_rel2meta[prom_name]
# Compute allprocs_prom2abs_list, abs2prom
allprocs_prom2abs_list[io][prom_name] = [abs_name]
abs2prom[io][abs_name] = prom_name
allprocs_abs2meta[abs_name] = {
meta_name: metadata[meta_name]
for meta_name in global_meta_names[io]
}
if is_input and 'src_indices' in metadata:
allprocs_abs2meta[abs_name]['has_src_indices'] = \
metadata['src_indices'] is not None
if metadata['add_input_src_indices'] and abs_name not in abs_in2prom_info:
# need a level for each system including '', so we don't
# subtract 1 from abs_in.split('.') which includes the var name
abs_in2prom_info[abs_name] = [None for s in abs_name.split('.')]
for prom_name, val in self._var_discrete[io].items():
abs_name = prefix + prom_name
# Compute allprocs_prom2abs_list, abs2prom
allprocs_prom2abs_list[io][prom_name] = [abs_name]
abs2prom[io][abs_name] = prom_name
# Compute allprocs_discrete (metadata for discrete vars)
self._var_allprocs_discrete[io][abs_name] = v = val.copy()
del v['val']
if self._var_discrete['input'] or self._var_discrete['output']:
self._discrete_inputs = _DictValues(self._var_discrete['input'])
self._discrete_outputs = _DictValues(self._var_discrete['output'])
else:
self._discrete_inputs = self._discrete_outputs = ()
self._serial_idxs = None
self._inconsistent_keys = set()
@collect_errors
def _setup_var_sizes(self):
"""
Compute the arrays of variable sizes for all variables/procs on this system.
"""
iproc = self.comm.rank
abs2idx = self._var_allprocs_abs2idx = {}
for io in ('input', 'output'):
sizes = self._var_sizes[io] = np.zeros((self.comm.size, len(self._var_rel_names[io])),
dtype=INT_DTYPE)
for i, (name, metadata) in enumerate(self._var_allprocs_abs2meta[io].items()):
sz = metadata['size']
sizes[iproc, i] = 0 if sz is None else sz
abs2idx[name] = i
if self.comm.size > 1:
my_sizes = sizes[iproc, :].copy()
self.comm.Allgather(my_sizes, sizes)
self._owned_sizes = self._var_sizes['output']
def _setup_partials(self):
"""
Process all partials and approximations that the user declared.
"""
self._subjacs_info = {}
self._jacobian = DictionaryJacobian(system=self)
self.setup_partials() # hook for component writers to specify sparsity patterns
# check to make sure that if num_par_fd > 1 that this system is actually doing FD.
# Unfortunately we have to do this check after system setup has been called because that's
# when declare_partials generally happens, so we raise an exception here instead of just
# resetting the value of num_par_fd (because the comm has already been split and possibly
# used by the system setup).
orig_comm = self._full_comm if self._full_comm is not None else self.comm
if self._num_par_fd > 1 and orig_comm.size > 1 and not (self._owns_approx_jac or
self._approx_schemes):
raise RuntimeError("%s: num_par_fd is > 1 but no FD is active." % self.msginfo)
for key, dct in self._declared_partials.items():
of, wrt = key
self._declare_partials(of, wrt, dct)
if self.matrix_free and self._subjacs_info:
issue_warning("matrix free component has declared the following "
f"partials: {sorted(self._subjacs_info)}, which will allocate "
"(possibly unnecessary) memory for each of those sub-jacobians.",
prefix=self.msginfo, category=DerivativesWarning)
[docs] def setup_partials(self):
"""
Declare partials.
This is meant to be overridden by component classes. All partials should be
declared here since this is called after all size/shape information is known for
all variables.
"""
pass
@property
def checking(self):
"""
Return True if check_partials or check_totals is executing.
Returns
-------
bool
True if we're running within check_partials or check_totals.
"""
return self._problem_meta is not None and self._problem_meta['checking']
def _run_root_only(self):
"""
Return the value of the run_root_only option and check for possible errors.
Returns
-------
bool
True if run_root_only is active.
"""
if self.options['run_root_only']:
if self.comm.size > 1 or (self._full_comm is not None and self._full_comm.size > 1):
if self._has_distrib_vars:
raise RuntimeError(f"{self.msginfo}: Can't set 'run_root_only' option when "
"a component has distributed variables.")
if self._num_par_fd > 1:
raise RuntimeError(f"{self.msginfo}: Can't set 'run_root_only' option when "
"using parallel FD.")
if self._problem_meta['using_par_deriv_color']:
raise RuntimeError(f"{self.msginfo}: Can't set 'run_root_only' option when "
"using parallel_deriv_color.")
return True
return False
def _update_wrt_matches(self, info):
"""
Determine the list of wrt variables that match the wildcard(s) given in declare_coloring.
Parameters
----------
info : dict
Coloring metadata dict.
"""
_, allwrt = self._get_partials_varlists()
wrt_patterns = info['wrt_patterns']
if wrt_patterns is None or '*' in wrt_patterns:
info['wrt_matches_rel'] = None
info['wrt_matches'] = None
return
matches_rel = set()
for w in wrt_patterns:
matches_rel.update(find_matches(w, allwrt))
# error if nothing matched
if not matches_rel:
raise ValueError("{}: Invalid 'wrt' variable(s) specified for colored approx partial "
"options: {}.".format(self.msginfo, wrt_patterns))
info['wrt_matches_rel'] = matches_rel
info['wrt_matches'] = [rel_name2abs_name(self, n) for n in matches_rel]
def _update_subjac_sparsity(self, sparsity):
"""
Update subjac sparsity info based on the given coloring.
The sparsity of the partial derivatives in this component will be used when computing
the sparsity of the total jacobian for the entire model. Without this, all of this
component's partials would be treated as dense, resulting in an overly conservative
coloring of the total jacobian.
Parameters
----------
sparsity : dict
A nested dict of the form dct[of][wrt] = (rows, cols, shape)
"""
# sparsity uses relative names, so we need to convert to absolute
prefix = self.pathname + '.' if self.pathname else None
for of, sub in sparsity.items():
if prefix:
of = prefix + of
for wrt, tup in sub.items():
if prefix:
wrt = prefix + wrt
abs_key = (of, wrt)
if abs_key in self._subjacs_info:
# add sparsity info to existing partial info
self._subjacs_info[abs_key]['sparsity'] = tup
[docs] def add_output(self, name, val=1.0, shape=None, units=None, res_units=None, desc='',
lower=None, upper=None, ref=1.0, ref0=0.0, res_ref=None, tags=None,
shape_by_conn=False, copy_shape=None, distributed=None):
"""
Add an output variable to the component.
Parameters
----------
name : str
Name of the variable in this component's namespace.
val : float or list or tuple or ndarray
The initial value of the variable being added in user-defined units. Default is 1.0.
shape : int or tuple or list or None
Shape of this variable, only required if val is not an array.
Default is None.
units : str or None
Units in which the output variables will be provided to the component during execution.
Default is None, which means it has no units.
res_units : str or None
Units in which the residuals of this output will be given to the user when requested.
Default is None, which means it has no units.
desc : str
Description of the variable.
lower : float or list or tuple or ndarray or Iterable or None
Lower bound(s) in user-defined units. It can be (1) a float, (2) an array_like
consistent with the shape arg (if given), or (3) an array_like matching the shape of
val, if val is array_like. A value of None means this output has no lower bound.
Default is None.
upper : float or list or tuple or ndarray or or Iterable None
Upper bound(s) in user-defined units. It can be (1) a float, (2) an array_like
consistent with the shape arg (if given), or (3) an array_like matching the shape of
val, if val is array_like. A value of None means this output has no upper bound.
Default is None.
ref : float or ndarray
Scaling parameter. The value in the user-defined units of this output variable when
the scaled value is 1. Default is 1.
ref0 : float or ndarray
Scaling parameter. The value in the user-defined units of this output variable when
the scaled value is 0. Default is 0.
res_ref : float or ndarray
Scaling parameter. The value in the user-defined res_units of this output's residual
when the scaled value is 1. Default is 1.
tags : str or list of strs or set of strs
User defined tags that can be used to filter what gets listed when calling
list_inputs and list_outputs.
shape_by_conn : bool
If True, shape this output to match its connected input(s).
copy_shape : str or None
If a str, that str is the name of a variable. Shape this output to match that of
the named variable.
distributed : bool
If True, this variable is a distributed variable, so it can have different sizes/values
across MPI processes.
Returns
-------
dict
Metadata for added variable.
"""
global _allowed_types
# First, type check all arguments
if (shape_by_conn or copy_shape) and (shape is not None or ndim(val) > 0):
raise ValueError("%s: If shape is to be set dynamically using 'shape_by_conn' or "
"'copy_shape', 'shape' and 'val' should be scalar, "
"but shape of '%s' and val of '%s' was given for variable '%s'."
% (self.msginfo, shape, val, name))
if not isinstance(name, str):
raise TypeError('%s: The name argument should be a string.' % self.msginfo)
if not _valid_var_name(name):
raise NameError("%s: '%s' is not a valid output name." % (self.msginfo, name))
if shape is not None and not isinstance(shape, (int, tuple, list, np.integer)):
raise TypeError("%s: The shape argument should be an int, tuple, or list but "
"a '%s' was given" % (self.msginfo, type(shape)))
if res_units is not None:
if not isinstance(res_units, str):
msg = '%s: The res_units argument should be a str or None' % self.msginfo
raise TypeError(msg)
res_units = simplify_unit(res_units, msginfo=self.msginfo)
if units is not None:
if not isinstance(units, str):
raise TypeError('%s: The units argument should be a str or None' % self.msginfo)
units = simplify_unit(units, msginfo=self.msginfo)
if tags is not None and not isinstance(tags, (str, set, list)):
raise TypeError('The tags argument should be a str, set, or list')
if not (copy_shape or shape_by_conn):
if not isscalar(val) and not isinstance(val, _allowed_types):
msg = '%s: The val argument should be a float, list, tuple, ndarray or Iterable'
raise TypeError(msg % self.msginfo)
# value, shape: based on args, making sure they are compatible
val, shape = ensure_compatible(name, val, shape)
if lower is not None:
lower = ensure_compatible(name, lower, shape)[0]
self._has_bounds = True
if upper is not None:
upper = ensure_compatible(name, upper, shape)[0]
self._has_bounds = True
# All refs: check the shape if necessary
for item, item_name in zip([ref, ref0, res_ref], ['ref', 'ref0', 'res_ref']):
if item is not None and not isscalar(item):
if not isinstance(item, _allowed_types):
raise TypeError(f'{self.msginfo}: The {item_name} argument should be a '
'float, list, tuple, ndarray or Iterable')
it = atleast_1d(item)
if it.shape != shape:
raise ValueError(f"{self.msginfo}: When adding output '{name}', expected "
f"shape {shape} but got shape {it.shape} for argument "
f"'{item_name}'.")
if isscalar(ref):
self._has_output_scaling |= ref != 1.0
else:
self._has_output_scaling |= np.any(ref != 1.0)
if isscalar(ref0):
self._has_output_scaling |= ref0 != 0.0
self._has_output_adder |= ref0 != 0.0
else:
self._has_output_scaling |= np.any(ref0)
self._has_output_adder |= np.any(ref0)
if isscalar(res_ref):
self._has_resid_scaling |= res_ref != 1.0
else:
self._has_resid_scaling |= np.any(res_ref != 1.0)
# until we get rid of component level distributed option, handle the case where
# component distributed has been set to True but variable distributed has been set
# to False by the caller.
if distributed is not False:
if distributed is None:
distributed = False
# using ._dict below to avoid tons of deprecation warnings
distributed = distributed or self.options._dict['distributed']['val']
metadata = {}
metadata.update({
'val': val,
'shape': shape,
'size': shape_to_len(shape),
'units': units,
'res_units': res_units,
'desc': desc,
'distributed': distributed,
'tags': make_set(tags),
'ref': format_as_float_or_array('ref', ref, flatten=True),
'ref0': format_as_float_or_array('ref0', ref0, flatten=True),
'res_ref': format_as_float_or_array('res_ref', res_ref, flatten=True, val_if_none=None),
'lower': lower,
'upper': upper,
'shape_by_conn': shape_by_conn,
'copy_shape': copy_shape
})
# this will get reset later if comm size is 1
self._has_distrib_vars |= metadata['distributed']
self._has_distrib_outputs |= metadata['distributed']
# We may not know the pathname yet, so we have to use name for now, instead of abs_name.
if self._static_mode:
var_rel2meta = self._static_var_rel2meta
var_rel_names = self._static_var_rel_names
else:
var_rel2meta = self._var_rel2meta
var_rel_names = self._var_rel_names
# Disallow dupes
if name in var_rel2meta:
raise ValueError("{}: Variable name '{}' already exists.".format(self.msginfo, name))
var_rel2meta[name] = metadata
var_rel_names['output'].append(name)
self._var_added(name)
return metadata
[docs] def add_discrete_output(self, name, val, desc='', tags=None):
"""
Add an output variable to the component.
Parameters
----------
name : str
Name of the variable in this component's namespace.
val : a picklable object
The initial value of the variable being added.
desc : str
Description of the variable.
tags : str or list of strs or set of strs
User defined tags that can be used to filter what gets listed when calling
list_inputs and list_outputs.
Returns
-------
dict
Metadata for added variable.
"""
if not isinstance(name, str):
raise TypeError('%s: The name argument should be a string.' % self.msginfo)
if not _valid_var_name(name):
raise NameError("%s: '%s' is not a valid output name." % (self.msginfo, name))
if tags is not None and not isinstance(tags, (str, set, list)):
raise TypeError('%s: The tags argument should be a str, set, or list' % self.msginfo)
metadata = {}
metadata.update({
'val': val,
'type': type(val),
'desc': desc,
'tags': make_set(tags)
})
if metadata['type'] == np.ndarray:
metadata.update({'shape': val.shape})
if self._static_mode:
var_rel2meta = self._static_var_rel2meta
else:
var_rel2meta = self._var_rel2meta
# Disallow dupes
if name in var_rel2meta:
raise ValueError("{}: Variable name '{}' already exists.".format(self.msginfo, name))
var_rel2meta[name] = self._var_discrete['output'][name] = metadata
self._var_added(name)
return metadata
def _var_added(self, name):
"""
Notify config that a variable has been added to this Component.
Parameters
----------
name : str
Name of the added variable.
"""
if self._problem_meta is not None and self._problem_meta['config_info'] is not None:
self._problem_meta['config_info']._var_added(self.pathname, name)
def _update_dist_src_indices(self, abs_in2out, all_abs2meta, all_abs2idx, all_sizes):
"""
Set default src_indices for any distributed inputs where they aren't set.
Parameters
----------
abs_in2out : dict
Mapping of connected inputs to their source. Names are absolute.
all_abs2meta : dict
Mapping of absolute names to metadata for all variables in the model.
all_abs2idx : dict
Dictionary mapping an absolute name to its allprocs variable index.
all_sizes : dict
Mapping of types to sizes of each variable in all procs.
Returns
-------
list
Names of inputs where src_indices were added.
"""
iproc = self.comm.rank
abs2meta_in = self._var_abs2meta['input']
all_abs2meta_in = all_abs2meta['input']
all_abs2meta_out = all_abs2meta['output']
abs_in2prom_info = self._problem_meta['abs_in2prom_info']
sizes_in = self._var_sizes['input']
sizes_out = all_sizes['output']
added_src_inds = []
# loop over continuous inputs
for i, (iname, meta_in) in enumerate(abs2meta_in.items()):
if meta_in['src_indices'] is None and iname not in abs_in2prom_info:
src = abs_in2out[iname]
dist_in = meta_in['distributed']
dist_out = all_abs2meta_out[src]['distributed']
if dist_in or dist_out:
gsize_out = all_abs2meta_out[src]['global_size']
gsize_in = all_abs2meta_in[iname]['global_size']
vout_sizes = sizes_out[:, all_abs2idx[src]]
offset = None
if gsize_out == gsize_in or (not dist_out and np.sum(vout_sizes)
== gsize_in):
# This assumes one of:
# 1) a distributed output with total size matching the total size of a
# distributed input
# 2) a non-distributed output with local size matching the total size of a
# distributed input
# 3) a non-distributed output with total size matching the total size of a
# distributed input
if dist_in:
offset = np.sum(sizes_in[:iproc, i])
end = offset + sizes_in[iproc, i]
else:
if src.startswith('_auto_ivc.'):
nzs = np.nonzero(vout_sizes)[0]
if nzs.size == 1:
# special case where we have a 'distributed' auto_ivc output
# that has a nonzero value in only one proc, so we can treat
# it like a non-distributed output. This happens in cases
# where an auto_ivc output connects to a variable that is
# remote on at least one proc.
offset = 0
end = vout_sizes[nzs[0]]
# total sizes differ and output is distributed, so can't determine mapping
if offset is None:
self._collect_error(f"{self.msginfo}: Can't determine src_indices "
f"automatically for input '{iname}'. They must be "
"supplied manually.", ident=(self.pathname, iname))
continue
if dist_in and not dist_out:
src_shape = self._get_full_dist_shape(src, all_abs2meta_out[src]['shape'])
else:
src_shape = all_abs2meta_out[src]['global_shape']
if offset == end:
idx = np.zeros(0, dtype=INT_DTYPE)
else:
idx = slice(offset, end)
meta_in['src_indices'] = indexer(idx, flat_src=True, src_shape=src_shape)
meta_in['flat_src_indices'] = True
added_src_inds.append(iname)
return added_src_inds
def _approx_partials(self, of, wrt, method='fd', **kwargs):
"""
Inform the framework that the specified derivatives are to be approximated.
Parameters
----------
of : str or list of str
The name of the residual(s) that derivatives are being computed for.
May also contain a glob pattern.
wrt : str or list of str
The name of the variables that derivatives are taken with respect to.
This can contain the name of any input or output variable.
May also contain a glob pattern.
method : str
The type of approximation that should be used. Valid options include:
- 'fd': Finite Difference
**kwargs : dict
Keyword arguments for controlling the behavior of the approximation.
"""
pattern_matches = self._find_partial_matches(of, wrt)
self._has_approx = True
for of_bundle, wrt_bundle in product(*pattern_matches):
of_pattern, of_matches = of_bundle
wrt_pattern, wrt_matches = wrt_bundle
if not of_matches:
raise ValueError('{}: No matches were found for of="{}"'.format(self.msginfo,
of_pattern))
if not wrt_matches:
raise ValueError('{}: No matches were found for wrt="{}"'.format(self.msginfo,
wrt_pattern))
info = self._subjacs_info
for abs_key in abs_key_iter(self, of_matches, wrt_matches):
meta = info[abs_key]
meta['method'] = method
meta.update(kwargs)
info[abs_key] = meta
[docs] def declare_partials(self, of, wrt, dependent=True, rows=None, cols=None, val=None,
method='exact', step=None, form=None, step_calc=None, minimum_step=None):
"""
Declare information about this component's subjacobians.
Parameters
----------
of : str or list of str
The name of the residual(s) that derivatives are being computed for.
May also contain a glob pattern.
wrt : str or list of str
The name of the variables that derivatives are taken with respect to.
This can contain the name of any input or output variable.
May also contain a glob pattern.
dependent : bool(True)
If False, specifies no dependence between the output(s) and the
input(s). This is only necessary in the case of a sparse global
jacobian, because if 'dependent=False' is not specified and
declare_partials is not called for a given pair, then a dense
matrix of zeros will be allocated in the sparse global jacobian
for that pair. In the case of a dense global jacobian it doesn't
matter because the space for a dense subjac will always be
allocated for every pair.
rows : ndarray of int or None
Row indices for each nonzero entry. For sparse subjacobians only.
cols : ndarray of int or None
Column indices for each nonzero entry. For sparse subjacobians only.
val : float or ndarray of float or scipy.sparse
Value of subjacobian. If rows and cols are not None, this will
contain the values found at each (row, col) location in the subjac.
method : str
The type of approximation that should be used. Valid options include:
'fd': Finite Difference, 'cs': Complex Step, 'exact': use the component
defined analytic derivatives. Default is 'exact'.
step : float
Step size for approximation. Defaults to None, in which case the approximation
method provides its default value.
form : str
Form for finite difference, can be 'forward', 'backward', or 'central'. Defaults
to None, in which case the approximation method provides its default value.
step_calc : str
Step type for computing the size of the finite difference step. It can be 'abs' for
absolute, 'rel_avg' for a size relative to the absolute value of the vector input, or
'rel_element' for a size relative to each value in the vector input. In addition, it
can be 'rel_legacy' for a size relative to the norm of the vector. For backwards
compatibilty, it can be 'rel', which is now equivalent to 'rel_avg'. Defaults to None,
in which case the approximation method provides its default value.
minimum_step : float
Minimum step size allowed when using one of the relative step_calc options.
Returns
-------
dict
Metadata dict for the specified partial(s).
"""
try:
method_func = _supported_methods[method]
except KeyError:
msg = '{}: d({})/d({}): method "{}" is not supported, method must be one of {}'
raise ValueError(msg.format(self.msginfo, of, wrt, method, sorted(_supported_methods)))
# lists aren't hashable so convert to tuples
if isinstance(of, list):
of = tuple(of)
if isinstance(wrt, list):
wrt = tuple(wrt)
meta = self._declared_partials[of, wrt]
meta['dependent'] = dependent
# If only one of rows/cols is specified
if (rows is None) ^ (cols is None):
raise ValueError('{}: d({})/d({}): If one of rows/cols is specified, then '
'both must be specified.'.format(self.msginfo, of, wrt))
if dependent:
meta['val'] = val
if rows is not None:
rows = np.array(rows, dtype=INT_DTYPE, copy=False)
cols = np.array(cols, dtype=INT_DTYPE, copy=False)
# Check the length of rows and cols to catch this easy mistake and give a
# clear message.
if len(cols) != len(rows):
raise RuntimeError("{}: d({})/d({}): declare_partials has been called "
"with rows and cols, which should be arrays of equal length,"
" but rows is length {} while cols is length "
"{}.".format(self.msginfo, of, wrt, len(rows), len(cols)))
if rows.size > 0 and rows.min() < 0:
msg = '{}: d({})/d({}): row indices must be non-negative'
raise ValueError(msg.format(self.msginfo, of, wrt))
if cols.size > 0 and cols.min() < 0:
msg = '{}: d({})/d({}): col indices must be non-negative'
raise ValueError(msg.format(self.msginfo, of, wrt))
meta['rows'] = rows
meta['cols'] = cols
# Check for repeated rows/cols indices.
size = len(rows)
if size > 0:
coo = coo_matrix((np.ones(size, dtype=np.short), (rows, cols)))
dsize = coo.data.size
csc = coo.tocsc()
# csc adds values at duplicate indices together, so result will be that data
# size is less if there are duplicates
if csc.data.size < dsize:
coo = csc.tocoo()
del csc
inds = np.where(coo.data > 1.)
dups = list(zip(coo.row[inds], coo.col[inds]))
raise RuntimeError("{}: d({})/d({}): declare_partials has been called "
"with rows and cols that specify the following duplicate"
" subjacobian entries: {}.".format(self.msginfo, of, wrt,
sorted(dups)))
if method_func is not None:
# we're doing approximations
self._has_approx = True
meta['method'] = method
self._get_approx_scheme(method)
default_opts = method_func.DEFAULT_OPTIONS
else:
default_opts = ()
if step:
if 'step' in default_opts:
meta['step'] = step
else:
raise RuntimeError("{}: d({})/d({}): 'step' is not a valid option for "
"'{}'".format(self.msginfo, of, wrt, method))
if minimum_step is not None:
if 'minimum_step' in default_opts:
meta['minimum_step'] = minimum_step
else:
raise RuntimeError("{}: d({})/d({}): 'minimum_step' is not a valid option for "
"'{}'".format(self.msginfo, of, wrt, method))
if form:
if 'form' in default_opts:
meta['form'] = form
else:
raise RuntimeError("{}: d({})/d({}): 'form' is not a valid option for "
"'{}'".format(self.msginfo, of, wrt, method))
if step_calc:
if 'step_calc' in default_opts:
meta['step_calc'] = step_calc
else:
raise RuntimeError("{}: d({})/d({}): 'step_calc' is not a valid option "
"for '{}'".format(self.msginfo, of, wrt, method))
return meta
[docs] def declare_coloring(self,
wrt=_DEFAULT_COLORING_META['wrt_patterns'],
method=_DEFAULT_COLORING_META['method'],
form=None,
step=None,
per_instance=_DEFAULT_COLORING_META['per_instance'],
num_full_jacs=_DEFAULT_COLORING_META['num_full_jacs'],
tol=_DEFAULT_COLORING_META['tol'],
orders=_DEFAULT_COLORING_META['orders'],
perturb_size=_DEFAULT_COLORING_META['perturb_size'],
min_improve_pct=_DEFAULT_COLORING_META['min_improve_pct'],
show_summary=_DEFAULT_COLORING_META['show_summary'],
show_sparsity=_DEFAULT_COLORING_META['show_sparsity']):
"""
Set options for deriv coloring of a set of wrt vars matching the given pattern(s).
Parameters
----------
wrt : str or list of str
The name or names of the variables that derivatives are taken with respect to.
This can contain input names, output names, or glob patterns.
method : str
Method used to compute derivative: "fd" for finite difference, "cs" for complex step.
form : str
Finite difference form, can be "forward", "central", or "backward". Leave
undeclared to keep unchanged from previous or default value.
step : float
Step size for finite difference. Leave undeclared to keep unchanged from previous
or default value.
per_instance : bool
If True, a separate coloring will be generated for each instance of a given class.
Otherwise, only one coloring for a given class will be generated and all instances
of that class will use it.
num_full_jacs : int
Number of times to repeat partial jacobian computation when computing sparsity.
tol : float
Tolerance used to determine if an array entry is nonzero during sparsity determination.
orders : int
Number of orders above and below the tolerance to check during the tolerance sweep.
perturb_size : float
Size of input/output perturbation during generation of sparsity.
min_improve_pct : float
If coloring does not improve (decrease) the number of solves more than the given
percentage, coloring will not be used.
show_summary : bool
If True, display summary information after generating coloring.
show_sparsity : bool
If True, display sparsity with coloring info after generating coloring.
"""
super().declare_coloring(wrt, method, form, step, per_instance,
num_full_jacs,
tol, orders, perturb_size, min_improve_pct,
show_summary, show_sparsity)
# create approx partials for all matches
meta = self.declare_partials('*', wrt, method=method, step=step, form=form)
meta['coloring'] = True
[docs] def set_check_partial_options(self, wrt, method='fd', form=None, step=None, step_calc=None,
minimum_step=None, directional=False):
"""
Set options that will be used for checking partial derivatives.
Parameters
----------
wrt : str or list of str
The name or names of the variables that derivatives are taken with respect to.
This can contain the name of any input or output variable.
May also contain a glob pattern.
method : str
Method for check: "fd" for finite difference, "cs" for complex step.
form : str
Finite difference form for check, can be "forward", "central", or "backward". Leave
undeclared to keep unchanged from previous or default value.
step : float
Step size for finite difference check. Leave undeclared to keep unchanged from previous
or default value.
step_calc : str
Step type for computing the size of the finite difference step. It can be 'abs' for
absolute, 'rel_avg' for a size relative to the absolute value of the vector input, or
'rel_element' for a size relative to each value in the vector input. In addition, it
can be 'rel_legacy' for a size relative to the norm of the vector. For backwards
compatibilty, it can be 'rel', which is now equivalent to 'rel_avg'. Defaults to None,
in which case the approximation method provides its default value..
minimum_step : float
Minimum step size allowed when using one of the relative step_calc options.
directional : bool
Set to True to perform a single directional derivative for each vector variable in the
pattern named in wrt.
"""
supported_methods = ('fd', 'cs')
if method not in supported_methods:
msg = "{}: Method '{}' is not supported, method must be one of {}"
raise ValueError(msg.format(self.msginfo, method, supported_methods))
if step and not isinstance(step, (int, float)):
msg = "{}: The value of 'step' must be numeric, but '{}' was specified."
raise ValueError(msg.format(self.msginfo, step))
supported_step_calc = ('abs', 'rel', 'rel_legacy', 'rel_avg', 'rel_element')
if step_calc and step_calc not in supported_step_calc:
msg = "{}: The value of 'step_calc' must be one of {}, but '{}' was specified."
raise ValueError(msg.format(self.msginfo, supported_step_calc, step_calc))
if not isinstance(wrt, (str, list, tuple)):
msg = "{}: The value of 'wrt' must be a string or list of strings, but a type " \
"of '{}' was provided."
raise ValueError(msg.format(self.msginfo, type(wrt).__name__))
if not isinstance(directional, bool):
msg = "{}: The value of 'directional' must be True or False, but a type " \
"of '{}' was provided."
raise ValueError(msg.format(self.msginfo, type(directional).__name__))
wrt_list = [wrt] if isinstance(wrt, str) else wrt
self._declared_partial_checks.append((wrt_list, method, form, step, step_calc,
minimum_step, directional))
def _get_check_partial_options(self):
"""
Return dictionary of partial options with pattern matches processed.
This is called by check_partials.
Returns
-------
dict(wrt: (options))
Dictionary keyed by name with tuples of options (method, form, step, step_calc,
minimum_step, directional)
"""
if not self._declared_partial_checks:
return {}
opts = {}
_, wrt = self._get_partials_varlists()
invalid_wrt = []
matrix_free = self.matrix_free
if matrix_free:
n_directional = 0
for data_tup in self._declared_partial_checks:
wrt_list, method, form, step, step_calc, minimum_step, directional = data_tup
for pattern in wrt_list:
matches = find_matches(pattern, wrt)
# if a non-wildcard var name was specified and not found, save for later Exception
if len(matches) == 0 and _valid_var_name(pattern):
invalid_wrt.append(pattern)
for match in matches:
if match in opts:
opt = opts[match]
# New assignments take precedence
keynames = ['method', 'form', 'step', 'step_calc', 'minimum_step',
'directional']
for name, value in zip(keynames,
[method, form, step, step_calc, minimum_step,
directional]):
if value is not None:
opt[name] = value
else:
opts[match] = {'method': method,
'form': form,
'step': step,
'step_calc': step_calc,
'minimum_step': minimum_step,
'directional': directional}
if matrix_free and directional:
n_directional += 1
if invalid_wrt:
msg = "{}: Invalid 'wrt' variables specified for check_partial options: {}."
raise ValueError(msg.format(self.msginfo, invalid_wrt))
if matrix_free:
if n_directional > 0 and n_directional < len(wrt):
msg = "{}: For matrix free components, directional should be set to True for " + \
"all inputs."
raise ValueError(msg.format(self.msginfo))
return opts
def _declare_partials(self, of, wrt, dct):
"""
Store subjacobian metadata for later use.
Parameters
----------
of : tuple of str
The names of the residuals that derivatives are being computed for.
May also contain glob patterns.
wrt : tuple of str
The names of the variables that derivatives are taken with respect to.
This can contain the name of any input or output variable.
May also contain glob patterns.
dct : dict
Metadata dict specifying shape, and/or approx properties.
"""
val = dct['val'] if 'val' in dct else None
is_scalar = isscalar(val)
dependent = dct['dependent']
if dependent:
if 'rows' in dct and dct['rows'] is not None: # sparse list format
rows = dct['rows']
cols = dct['cols']
if is_scalar:
val = np.full(rows.size, val, dtype=float)
is_scalar = False
elif val is not None:
# np.promote_types will choose the smallest dtype that can contain
# both arguments
val = atleast_1d(val)
safe_dtype = promote_types(val.dtype, float)
val = val.astype(safe_dtype, copy=False)
if rows.shape != val.shape:
raise ValueError('{}: d({})/d({}): If rows and cols are specified, val '
'must be a scalar or have the same shape, val: {}, '
'rows/cols: {}'.format(self.msginfo, of, wrt,
val.shape, rows.shape))
else:
val = np.zeros_like(rows, dtype=float)
if rows.size > 0:
rows_max = rows.max()
cols_max = cols.max()
else:
rows_max = cols_max = 0
else:
if val is not None and not is_scalar and not issparse(val):
val = atleast_2d(val)
val = val.astype(promote_types(val.dtype, float), copy=False)
rows_max = cols_max = 0
rows = None
cols = None
pattern_matches = self._find_partial_matches(of, '*' if wrt is None else wrt)
abs2meta_in = self._var_abs2meta['input']
abs2meta_out = self._var_abs2meta['output']
is_array = isinstance(val, ndarray)
patmeta = dict(dct)
patmeta_not_none = {k: v for k, v in dct.items() if v is not None}
for of_bundle, wrt_bundle in product(*pattern_matches):
of_pattern, of_matches = of_bundle
wrt_pattern, wrt_matches = wrt_bundle
if not of_matches:
raise ValueError('{}: No matches were found for of="{}"'.format(self.msginfo,
of_pattern))
if not wrt_matches:
raise ValueError('{}: No matches were found for wrt="{}"'.format(self.msginfo,
wrt_pattern))
for abs_key in abs_key_iter(self, of_matches, wrt_matches):
if not dependent:
if abs_key in self._subjacs_info:
del self._subjacs_info[abs_key]
continue
if abs_key in self._subjacs_info:
meta = self._subjacs_info[abs_key]
meta.update(patmeta_not_none)
else:
meta = patmeta.copy()
of, wrt = abs_key
meta['rows'] = rows
meta['cols'] = cols
csz = abs2meta_in[wrt]['size'] if wrt in abs2meta_in else abs2meta_out[wrt]['size']
meta['shape'] = shape = (abs2meta_out[of]['size'], csz)
dist_out = abs2meta_out[of]['distributed']
if wrt in abs2meta_in:
dist_in = abs2meta_in[wrt]['distributed']
else:
dist_in = abs2meta_out[wrt]['distributed']
if dist_in and not dist_out and not self.matrix_free:
rel_key = abs_key2rel_key(self, abs_key)
raise RuntimeError(f"{self.msginfo}: component has defined partial {rel_key} "
"which is a non-distributed output wrt a distributed input."
" This is only supported using the matrix free API.")
if shape[0] == 0 or shape[1] == 0:
msg = "{}: '{}' is an array of size 0"
if shape[0] == 0:
if dist_out:
# distributed vars are allowed to have zero size inputs on some procs
rows_max = -1
else:
# non-distributed vars are not allowed to have zero size inputs
raise ValueError(msg.format(self.msginfo, of))
if shape[1] == 0:
if not dist_in:
# non-distributed vars are not allowed to have zero size outputs
raise ValueError(msg.format(self.msginfo, wrt))
else:
# distributed vars are allowed to have zero size outputs on some procs
cols_max = -1
if val is None:
# we can only get here if rows is None (we're not sparse list format)
meta['val'] = np.zeros(shape)
elif is_array:
if rows is None and val.shape != shape and val.size == shape[0] * shape[1]:
meta['val'] = val = val.copy().reshape(shape)
else:
meta['val'] = val.copy()
elif is_scalar:
meta['val'] = np.full(shape, val, dtype=float)
else:
meta['val'] = val
if rows_max >= shape[0] or cols_max >= shape[1]:
of, wrt = abs_key2rel_key(self, abs_key)
raise ValueError(f"{self.msginfo}: d({of})/d({wrt}): Expected {shape[0]}x"
f"{shape[1]} but declared at least {rows_max + 1}x"
f"{cols_max + 1}")
self._check_partials_meta(abs_key, meta['val'],
shape if rows is None else (rows.shape[0], 1))
self._subjacs_info[abs_key] = meta
def _find_partial_matches(self, of_pattern, wrt_pattern, use_resname=False):
"""
Find all partial derivative matches from of and wrt.
Parameters
----------
of_pattern : str or list of str
The relative name of the residual(s) that derivatives are being computed for.
May also contain a glob pattern.
wrt_pattern : str or list of str
The relative name of the variables that derivatives are taken with respect to.
This can contain the name of any input or output variable.
May also contain a glob pattern.
use_resname : bool
If True, use residual names for 'of' patterns.
Returns
-------
tuple(list, list)
Pair of lists containing pattern matches (if any). Returns (of_matches, wrt_matches)
where of_matches is a list of tuples (pattern, matches) and wrt_matches is a list of
tuples (pattern, output_matches, input_matches).
"""
of_list = [of_pattern] if isinstance(of_pattern, str) else of_pattern
wrt_list = [wrt_pattern] if isinstance(wrt_pattern, str) else wrt_pattern
ofs, wrts = self._get_partials_varlists(use_resname=use_resname)
of_pattern_matches = [(pattern, find_matches(pattern, ofs)) for pattern in of_list]
wrt_pattern_matches = [(pattern, find_matches(pattern, wrts)) for pattern in wrt_list]
return of_pattern_matches, wrt_pattern_matches
def _check_partials_meta(self, abs_key, val, shape):
"""
Check a given partial derivative and metadata for the correct shapes.
Parameters
----------
abs_key : tuple(str, str)
The of/wrt pair (given absolute names) defining the partial derivative.
val : ndarray
Subjac value.
shape : tuple
Expected shape of val.
"""
out_size, in_size = shape
if in_size == 0 and self.comm.rank != 0: # 'inactive' component
return
if val is not None:
val_shape = val.shape
if len(val_shape) == 1:
val_out, val_in = val_shape[0], 1
else:
val_out, val_in = val.shape
if val_out > out_size or val_in > in_size:
of, wrt = abs_key2rel_key(self, abs_key)
msg = '{}: d({})/d({}): Expected {}x{} but val is {}x{}'
raise ValueError(msg.format(self.msginfo, of, wrt, out_size, in_size,
val_out, val_in))
def _set_approx_partials_meta(self):
"""
Add approximations for those partials registered with method=fd or method=cs.
"""
self._get_static_wrt_matches()
subjacs = self._subjacs_info
wrtset = set()
subjac_keys = self._get_approx_subjac_keys()
# go through subjac keys in reverse and only add approx for the last of each wrt
# (this prevents warnings that could confuse users)
for i in range(len(subjac_keys) - 1, -1, -1):
key = subjac_keys[i]
if key[1] not in wrtset:
wrtset.add(key[1])
meta = subjacs[key]
self._approx_schemes[meta['method']].add_approximation(key, self, meta)
def _guess_nonlinear(self):
"""
Provide initial guess for states.
Does nothing on any non-implicit component.
"""
pass
def _clear_iprint(self):
"""
Clear out the iprint stack from the solvers.
Components don't have nested solvers, so do nothing to prevent errors.
"""
pass
def _check_first_linearize(self):
if self._first_call_to_linearize:
self._first_call_to_linearize = False # only do this once
if coloring_mod._use_partial_sparsity:
coloring = self._get_coloring()
if coloring is not None:
if not self._coloring_info['dynamic']:
coloring._check_config_partial(self)
self._update_subjac_sparsity(coloring.get_subjac_sparsity())
self._jacobian._restore_approx_sparsity()
def _resolve_src_inds(self):
abs2prom = self._var_abs2prom['input']
abs_in2prom_info = self._problem_meta['abs_in2prom_info']
all_abs2meta_in = self._var_allprocs_abs2meta['input']
abs2meta_in = self._var_abs2meta['input']
conns = self._problem_meta['model_ref']()._conn_global_abs_in2out
all_abs2meta_out = self._problem_meta['model_ref']()._var_allprocs_abs2meta['output']
for tgt, meta in abs2meta_in.items():
if tgt in abs_in2prom_info:
pinfo = abs_in2prom_info[tgt][-1] # component always last in the plist
if pinfo is not None:
inds, flat, shape = pinfo
if inds is None:
if meta['add_input_src_indices']:
if shape is None:
shape = all_abs2meta_out[conns[tgt]]['global_shape']
meta['src_shape'] = shape
inds = meta['src_indices']
else:
all_abs2meta_in[tgt]['has_src_indices'] = True
meta['src_shape'] = shape = all_abs2meta_out[conns[tgt]]['global_shape']
if meta['add_input_src_indices']:
inds = convert_src_inds(inds, shape, meta['src_indices'], shape)
elif inds._flat_src:
meta['flat_src_indices'] = True
elif meta['flat_src_indices'] is None:
meta['flat_src_indices'] = flat
if inds is not None:
try:
if not isinstance(inds, Indexer):
meta['src_indices'] = inds = indexer(inds, flat_src=flat,
src_shape=shape)
else:
meta['src_indices'] = inds = inds.copy()
inds.set_src_shape(shape)
self._var_prom2inds[abs2prom[tgt]] = [shape, inds, flat]
except Exception:
type_exc, exc, tb = sys.exc_info()
self._collect_error(f"When accessing '{conns[tgt]}' with src_shape "
f"{shape} from '{pinfo.prom_path()}' using "
f"src_indices {inds}: {exc}", exc_type=type_exc,
tback=tb, ident=(conns[tgt], tgt))
elif meta['add_input_src_indices']:
self._var_prom2inds[abs2prom[tgt]] = [meta['shape'], meta['src_indices'],
meta['src_indices']._flat_src]
def _check_consistent_serial_dinputs(self, nz_dist_outputs):
"""
Check consistency across ranks for serial d_inputs variables.
This is used primarily to test that `compute_jacvec_product` and `apply_linear` methods
follow the OpenMDAO convention that in reverse mode, the component should perform
'allreduce' sorts of operations only for derivatives of distributed outputs with-respect-to
serial inputs. This should result in serial input derivatives being consistent across all
ranks in the Component's communicator.
Parameters
----------
nz_dist_outputs : set or list
Set of distributed outputs with nonzero values for the most recent _apply_linear call.
"""
if not self.checking or not self._has_distrib_outputs or self.comm.size < 2:
return
if self._serial_idxs is None:
ranges = defaultdict(list)
output_len = 0 if self.is_explicit() else len(self._outputs)
for name, offset, end, vec, slc, dist_sizes in self._jac_wrt_iter():
if dist_sizes is None: # not distributed
if offset != end:
if vec is self._outputs:
ranges[vec].append(range(offset, end))
else:
ranges[vec].append(range(offset - output_len, end - output_len))
self._serial_idxs = []
for vec, rlist in ranges.items():
if rlist:
self._serial_idxs.append((vec, np.hstack(rlist)))
for vec, inds in self._serial_idxs:
# _jac_wrt_iter gives us _input and possibly _output vecs (for implicit comps), but we
# want to check _dinputs and _doutputs
v = self._dinputs if vec is self._inputs else self._doutputs
result = inconsistent_across_procs(self.comm, v.asarray()[inds])
if self.comm.rank == 0 and np.any(result):
bad_inds = np.arange(len(v), dtype=INT_DTYPE)[inds][result]
bad_mask = np.zeros(len(v), dtype=bool)
bad_mask[bad_inds] = True
for inname, slc in v.get_slice_dict().items():
if np.any(bad_mask[slc]):
for outname in nz_dist_outputs:
key = (outname, inname)
self._inconsistent_keys.add(key)
def _get_dist_nz_dresids(self):
"""
Get names of distributed resids that are non-zero prior to computing derivatives.
This should only be called when 'rev' mode is active.
Returns
-------
list of str
List of names of distributed resids that have nonzero entries.
"""
nzresids = []
dresids = self._dresiduals.asarray()
for of, start, end, _full_slice, dist_sizes in self._jac_of_iter():
if dist_sizes is not None:
if np.any(dresids[start:end]):
nzresids.append(of)
full_nzresids = set()
if self.comm.rank == 0:
for nzoutlist in self.comm.gather(nzresids, root=0):
full_nzresids.update(nzoutlist)
return full_nzresids
self.comm.gather(nzresids, root=0)
return nzresids
def _has_fast_rel_lookup(self):
"""
Return True if this System should have fast relative variable name lookup in vectors.
Returns
-------
bool
True if this System should have fast relative variable name lookup in vectors.
"""
return True
class _DictValues(object):
"""
A dict-like wrapper for a dict of metadata, where getitem returns 'val' from metadata.
"""
def __init__(self, dct):
self._dict = dct
def __getitem__(self, key):
return self._dict[key]['val']
def __setitem__(self, key, value):
self._dict[key]['val'] = value
def __contains__(self, key):
return key in self._dict
def __len__(self):
return len(self._dict)
def items(self):
return [(key, self._dict[key]['val']) for key in self._dict]
