# -*- coding: utf-8 -*-
# Copyright 2021 IRT Saint Exupéry, https://www.irt-saintexupery.com
#
# This program is free software; you can redistribute it and/or
# modify it under the terms of the GNU Lesser General Public
# License version 3 as published by the Free Software Foundation.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
# Lesser General Public License for more details.
#
# You should have received a copy of the GNU Lesser General Public License
# along with this program; if not, write to the Free Software Foundation,
# Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
# Contributors:
# INITIAL AUTHORS - initial API and implementation and/or initial
# documentation
# :author: Francois Gallard
# OTHER AUTHORS - MACROSCOPIC CHANGES
"""
Abstraction of processes
************************
"""
from __future__ import absolute_import, division, unicode_literals
import inspect
import sys
from collections import defaultdict
from copy import deepcopy
from multiprocessing import Value, cpu_count
from multiprocessing.sharedctypes import Synchronized
from os.path import abspath, dirname, join
from timeit import default_timer as timer
from future import standard_library
from numpy import concatenate, empty, zeros
from six import string_types
from gemseo.caches.cache_factory import CacheFactory
from gemseo.core.grammar import InvalidDataException, SimpleGrammar
from gemseo.core.jacobian_assembly import JacobianAssembly
from gemseo.core.json_grammar import JSONGrammar
from gemseo.utils.derivatives_approx import EPSILON, DisciplineJacApprox
try:
import cPickle as pickle
except ImportError:
import pickle
standard_library.install_aliases()
from gemseo import LOGGER
[docs]def default_dict_factory():
"""Instantiate a defaultdict(None) object."""
return defaultdict(None)
[docs]class MDODiscipline(object):
"""A software integrated in the workflow.
The inputs and outputs are defined in a grammar, which can be
either a SimpleGrammar or a JSONGrammar, or your own which
derives from the Grammar abstract class
To be used, use a subclass and implement the _run method
which defined the execution of the software.
Typically, in the _run method, get the inputs from the
input grammar, call your software, and write the outputs
to the output grammar.
The JSON Grammar are automatically detected when in the same
folder as your subclass module and named "CLASSNAME_input.json"
use auto_detect_grammar_files=True to activate this option
"""
STATUS_VIRTUAL = "VIRTUAL"
STATUS_PENDING = "PENDING"
STATUS_DONE = "DONE"
STATUS_RUNNING = "RUNNING"
STATUS_FAILED = "FAILED"
JSON_GRAMMAR_TYPE = "JSON"
SIMPLE_GRAMMAR_TYPE = "Simple"
COMPLEX_STEP = "complex_step"
FINITE_DIFFERENCES = "finite_differences"
SIMPLE_CACHE = "SimpleCache"
HDF5_CACHE = "HDF5Cache"
MEMORY_FULL_CACHE = "MemoryFullCache"
APPROX_MODES = [FINITE_DIFFERENCES, COMPLEX_STEP]
AVAILABLE_MODES = (
JacobianAssembly.AUTO_MODE,
JacobianAssembly.DIRECT_MODE,
JacobianAssembly.ADJOINT_MODE,
JacobianAssembly.REVERSE_MODE,
FINITE_DIFFERENCES,
COMPLEX_STEP,
)
RE_EXECUTE_DONE_POLICY = "RE_EXEC_DONE"
RE_EXECUTE_NEVER_POLICY = "RE_EXEC_NEVER"
N_CPUS = cpu_count()
def __init__(
self,
name=None,
input_grammar_file=None,
output_grammar_file=None,
auto_detect_grammar_files=False,
grammar_type=JSON_GRAMMAR_TYPE,
cache_type=SIMPLE_CACHE,
cache_file_path=None,
):
"""Constructor.
:param name: the name of the discipline
:param input_grammar_file: the file for input grammar description,
if None, name + "_input.json" is used
:param output_grammar_file: the file for output grammar description,
if None, name + "_output.json" is used
:param auto_detect_grammar_files: if no input and output grammar files
are provided, auto_detect_grammar_files uses a naming convention
to associate a
grammar file to a discipline:
searches in the "comp_dir" directory containing the
discipline source file for files basenames
self.name _input.json and self.name _output.json
:param grammar_type: the type of grammar to use for IO declaration
either JSON_GRAMMAR_TYPE or SIMPLE_GRAMMAR_TYPE
:param cache_type: type of cache policy, SIMPLE_CACHE
or HDF5_CACHE
:param cache_file_path: the file to store the data,
mandatory when HDF caching is used
"""
self.input_grammar = None # : input grammar
self.output_grammar = None # : output grammar
self.grammar_type = grammar_type
self.comp_dir = None
# : data converters between execute and _run
self.data_processor = None
self._default_inputs = {} # : default data to be set if not passed
# Allow to re execute the same discipline twice, only if did not fail
# and not running
self.re_exec_policy = self.RE_EXECUTE_DONE_POLICY
# : list of outputs that shall be null, to be considered as residuals
self.residual_variables = []
self._differentiated_inputs = [] # : outputs to differentiate
# : inputs to be used for differenciation
self._differentiated_outputs = []
self._n_calls = None # : number of calls to execute()
self._exec_time = None # : cumulated execution time
# : number of calls to linearize()
self._n_calls_linearize = None
self._in_data_hash_dict = {}
self.jac = None # : Jacobians of outputs wrt inputs dictionary
# : True if linearize() has already been called
self._is_linearized = False
self._jac_approx = None # Jacobian approximation object
self._linearize_on_last_state = False # If true, the linearization
# is performed on the state computed by the disciplines
# (MDAs for instance) otherwise, the inputs that are also outputs
# are reset at the previous inputs value be fore calling
# _compute_jacobian
# the last execution was due to a linearization
self.exec_for_lin = False
if name is None:
self.name = self.__class__.__name__
else:
self.name = name
self.cache = None
self._cache_type = cache_type
self._cache_file_path = cache_file_path
self._cache_tolerance = 0.0
self._cache_hdf_node_name = None
self._cache_factory = CacheFactory()
# By default, dont use approximate cache
# It is up to the user to choose to optimize CPU time with this or not
self.set_cache_policy(cache_type=cache_type, cache_hdf_file=cache_file_path)
# linearize mode :auto, adjoint, direct
self._linearization_mode = JacobianAssembly.AUTO_MODE
self_module = sys.modules.get(self.__class__.__module__)
has_file = hasattr(self_module, "__file__")
if has_file:
f_class = inspect.getfile(self.__class__)
self.comp_dir = abspath(dirname(f_class))
if input_grammar_file is None and auto_detect_grammar_files:
input_grammar_file = self.auto_get_grammar_file(True)
if output_grammar_file is None and auto_detect_grammar_files:
output_grammar_file = self.auto_get_grammar_file(False)
self._instantiate_grammars(
input_grammar_file, output_grammar_file, grammar_type
)
self.local_data = {} # : the inputs and outputs data
# : The current status of execution
self._status = self.STATUS_PENDING
self._cache_was_loaded = False
self._init_shared_attrs()
self._status_observers = []
def _init_shared_attrs(self):
"""Initialize the shared attributes in multiprocessing."""
self._n_calls = Value("i", 0)
self._exec_time = Value("d", 0.0)
self._n_calls_linearize = Value("i", 0)
def __init_cache_attr(self):
"""Initialize cache attributes after deserialization."""
if self._cache_type == self.HDF5_CACHE:
self.cache = None
self.set_cache_policy(
self.HDF5_CACHE,
self._cache_tolerance,
self._cache_file_path,
self._cache_hdf_node_name,
)
@property
def n_calls(self):
"""Return the number of calls to execute() which triggered the _run().
Multiprocessing safe.
"""
return self._n_calls.value
@n_calls.setter
def n_calls(self, value):
"""Set the number of calls to execute() which triggered the _run().
Multiprocessing safe
:param value: the value of n_calls
"""
self._n_calls.value = value
@property
def exec_time(self):
"""Return the cumulated execution time.
Multiprocessing safe.
"""
return self._exec_time.value
@exec_time.setter
def exec_time(self, value):
"""Set the cumulated execution time.
Multiprocessing safe
:param value: the value of exec_time
"""
self._exec_time.value = value
@property
def n_calls_linearize(self):
"""Return the number of calls to linearize() which triggered the
_compute_jacobian() method.
Multiprocessing safe.
"""
return self._n_calls_linearize.value
@n_calls_linearize.setter
def n_calls_linearize(self, value):
"""Set the number of calls to linearize() which triggered the
_compute_jacobian() method
Multiprocessing safe
:param value: the value of n_calls_linearize
"""
self._n_calls_linearize.value = value
[docs] def auto_get_grammar_file(self, is_input=True, name=None, comp_dir=None):
"""Use a naming convention to associate a grammar file to a discipline.
This method searches in the "comp_dir" directory containing the
discipline source file for files basenames self.name _input.json and
self.name _output.json
:param is_input: if True, searches for _input.json,
otherwise _output.json (Default value = True)
:param name: the name of the discipline (Default value = None)
:param comp_dir: the containing directory
if None, use self.comp_dir (Default value = None)
:returns: path to the grammar file
:rtype: string
"""
if comp_dir is None:
comp_dir = self.comp_dir
if name is None:
name = self.name
if is_input:
endf = "_input.json"
else:
endf = "_output.json"
return join(comp_dir, name + endf)
[docs] def add_differentiated_outputs(self, outputs=None):
"""Add outputs to the differentiation list.
Update self._differentiated_inputs with inputs.
:param outputs: list of output variables to differentiate
if None, all outputs of discipline are used
"""
if (outputs is not None) and (not self.is_all_outputs_existing(outputs)):
raise ValueError(
"Cannot differentiate discipline "
+ self.name
+ " outputs that are not"
+ " among discipline outputs: "
+ str(self.get_output_data_names())
)
out_diff = self._differentiated_outputs
if outputs is None:
outputs = self.get_output_data_names()
self._differentiated_outputs = list(set(out_diff) | set(outputs))
[docs] def set_cache_policy(
self,
cache_type=SIMPLE_CACHE,
cache_tolerance=0.0,
cache_hdf_file=None,
cache_hdf_node_name=None,
):
"""Set the type of cache to use and the tolerance level.
This method set the cache policy to cache data whose inputs are close
to inputs whose outputs are already cached. The cache can be either a
simple cache recording the last execution or a full cache storing all
executions. Caching data can be either in-memory, e.g.
:class:`.SimpleCache` and
:class:`.MemoryFullCache` ,
or on the disk, e.g.
:class:`.HDF5Cache` .
:attr:`.CacheFactory.caches`
provides the list of available types of caches.
:param str cache_type: type of cache to use.
:param float cache_tolerance: tolerance for the approximate cache
maximal relative norm difference to consider that
two input arrays are equal
:param str cache_hdf_file: the file to store the data,
mandatory when HDF caching is used
:param str cache_hdf_node_name: name of the HDF
dataset to store the discipline
data. If None, self.name is used
"""
create_cache = self._cache_factory.create
if cache_type == self.HDF5_CACHE:
not_same_file = cache_hdf_file != self._cache_file_path
not_same_node = cache_hdf_node_name != self._cache_hdf_node_name
cache_none = self.cache is None
already_hdf = self._cache_type == self.HDF5_CACHE
not_already_hdf = self._cache_type != self.HDF5_CACHE
if cache_none or (
(already_hdf and (not_same_file or not_same_node)) or not_already_hdf
):
node_path = cache_hdf_node_name or self.name
self.cache = create_cache(
self.HDF5_CACHE,
hdf_file_path=cache_hdf_file,
hdf_node_path=node_path,
tolerance=cache_tolerance,
name=self.name,
)
self._cache_hdf_node_name = cache_hdf_node_name
self._cache_file_path = cache_hdf_file
elif cache_type != self._cache_type or self.cache is None:
self.cache = create_cache(
cache_type, tolerance=cache_tolerance, name=self.name
)
else:
LOGGER.warning(
"Cache policy is already set to %s. To clear the"
" discipline cache, use its clear() method.",
cache_type,
)
self._cache_type = cache_type
self._cache_tolerance = cache_tolerance
[docs] def get_sub_disciplines(self): # pylint: disable=R0201
"""Gets the sub disciplines of self
By default, empty
:returns: the list of disciplines
"""
return []
[docs] def get_expected_workflow(self):
"""Return the expected execution sequence.
This method is used for XDSM representation
Default to the execution of the discipline itself
See MDOFormulation.get_expected_workflow
"""
# avoid circular dependency
from gemseo.core.execution_sequence import ExecutionSequenceFactory
return ExecutionSequenceFactory.serial(self)
[docs] def get_expected_dataflow(self): # pylint: disable=R0201
"""Return the expected data exchange sequence.
This method is used for the XDSM representation.
Default to empty list
See MDOFormulation.get_expected_dataflow
:returns: a list representing the data exchange arcs
"""
return []
def _instantiate_grammars(
self, input_grammar_file, output_grammar_file, grammar_type=JSON_GRAMMAR_TYPE
):
"""Create the input and output grammars.
:param input_grammar_file: the input file of the grammar
:param output_grammar_file: the output file of the grammar
:param grammar_type: the type of grammar to use, JSON,
Simple or yours ! (Default value = JSON_GRAMMAR_TYPE)
"""
if grammar_type == self.JSON_GRAMMAR_TYPE:
self.input_grammar = JSONGrammar(
name=self.name + "_input",
schema_file=input_grammar_file,
grammar_type="input",
)
self.output_grammar = JSONGrammar(
name=self.name + "_output",
schema_file=output_grammar_file,
grammar_type="output",
)
elif grammar_type == self.SIMPLE_GRAMMAR_TYPE:
self.input_grammar = SimpleGrammar(name=self.name + "_input")
self.output_grammar = SimpleGrammar(name=self.name + "_output")
else:
raise ValueError("Unknown grammar type: " + str(grammar_type))
def _run(self):
"""Define the execution of the process, given that data has been checked.
To be overloaded by sub classes.
"""
raise NotImplementedError()
def _filter_inputs(self, input_data=None):
"""Load the input data and adds default data when not present.
This method filters the inputs that shall not be used by the
discipline.
:param input_data: a data dictionary (Default value = None)
:returns: the filtered data dictionary
"""
if input_data is None:
return deepcopy(self.default_inputs)
if not isinstance(input_data, dict):
raise TypeError(
"Input data must be a dict, got " + str(type(input_data)) + " instead"
)
# Take default inputs if not in input_data
filt_inputs = self._default_inputs.copy() # Shallow copy
filt_inputs.update(input_data)
# Remove inputs that should not be there
in_names = self.get_input_data_names()
filt_inputs = {key: val for key, val in filt_inputs.items() if key in in_names}
return filt_inputs
def _filter_local_data(self):
"""Filter the local data after execution.
This method removes data that are neither inputs nor outputs.
"""
all_data_names = self.get_input_output_data_names()
self.local_data = {
key: val for key, val in self.local_data.items() if key in all_data_names
}
def _check_status_before_run(self):
"""Check the status of the discipline before calling _run.
Check the status of the discipline depending on self.re_execute_policy.
if self.re_exec_policy == RE_EXECUTE_NEVER_POLICY:
status shall be either PENDING or VIRTUAL
if self.re_exec_policy == RE_EXECUTE_NEVER_POLICY:
if status is DONE, self.reset_statuses_for_run is called prior run
otherwise status must be VIRTUAL or PENDING
"""
status_ok = True
if self.status == self.STATUS_RUNNING:
status_ok = False
if self.re_exec_policy == self.RE_EXECUTE_NEVER_POLICY:
if self.status not in [self.STATUS_PENDING, self.STATUS_VIRTUAL]:
status_ok = False
elif self.re_exec_policy == self.RE_EXECUTE_DONE_POLICY:
if self.status == self.STATUS_DONE:
self.reset_statuses_for_run()
status_ok = True
elif self.status not in [self.STATUS_PENDING, self.STATUS_VIRTUAL]:
status_ok = False
else:
raise ValueError("Unknown re_exec_policy :" + str(self.re_exec_policy))
if not status_ok:
raise ValueError(
"Trying to run a discipline "
+ str(type(self))
+ " with status: "
+ str(self.status)
+ " while re_exec_policy is : "
+ str(self.re_exec_policy)
)
def __get_input_data_for_cache(self, input_data, in_names):
"""
Prepares the input data dict for caching
:param input_data: input data dict
:param in_names: input data names
:returns: a data dict
"""
in_and_out = set(in_names) & set(self.get_output_data_names())
cached_inputs = dict(input_data.items())
for key in in_and_out:
val = input_data.get(key)
if val is not None:
# If also an output, keeps a copy of the original input value
cached_inputs[key] = deepcopy(val)
return cached_inputs
[docs] def execute(self, input_data=None):
"""Execute the discipline.
This method executes the discipline:
* Adds default inputs to the input_data if some inputs are not defined
in input_data but exist in self._default_data
* Checks if the last execution of the discipline wan not called with
identical inputs, cached in self.cache, if yes, directly
return self.cache.get_output_cache(inputs)
* Caches the inputs
* Checks the input data against self.input_grammar
* if self.data_processor is not None: runs the preprocessor
* updates the status to RUNNING
* calls the _run() method, that shall be defined
* if self.data_processor is not None: runs the postprocessor
* checks the output data
* Caches the outputs
* updates the status to DONE or FAILED
* updates summed execution time
:param input_data: the input data dict needed to execute the
disciplines according to the discipline input grammar
(Default value = None)
:type input_data: dict
:returns: the discipline local data after execution
:rtype: dict
"""
# Load the default_inputs if the user did not provide all required data
input_data = self._filter_inputs(input_data)
# Check if the cache already the contains outputs associated to these
# inputs
in_names = self.get_input_data_names()
out_cached, out_jac = self.cache.get_outputs(input_data, in_names)
if out_cached is not None:
self.__update_local_data_from_cache(input_data, out_cached, out_jac)
return self.local_data
# Cache was not loaded, see self.linearize
self._cache_was_loaded = False
# Save the state of the inputs
cached_inputs = self.__get_input_data_for_cache(input_data, in_names)
self._check_status_before_run()
self.check_input_data(input_data)
self.local_data = {}
self.local_data.update(input_data)
processor = self.data_processor
# If the data processor is set, pre-process the data before _run
# See gemseo.core.data_processor module
if processor is not None:
self.local_data = processor.pre_process_data(input_data)
self.status = self.STATUS_RUNNING
self._is_linearized = False
self.__increment_n_calls()
t_0 = timer()
try:
# Effectively run the discipline, the _run method has to be
# Defined by the subclasses
self._run()
except Exception:
self.status = self.STATUS_FAILED
# Update the status but
# raise the same exception
raise
self.__increment_exec_time(t_0)
self.status = self.STATUS_DONE
# If the data processor is set, post process the data after _run
# See gemseo.core.data_processor module
if processor is not None:
self.local_data = processor.post_process_data(self.local_data)
# Filter data that is neither outputs nor inputs
self._filter_local_data()
self.check_output_data()
# Caches output data in case the discipline is called twice in a row
# with the same inputs
out_names = self.get_output_data_names()
self.cache.cache_outputs(cached_inputs, in_names, self.local_data, out_names)
# Some disciplines are always linearized during execution, cache the
# jac in this case
if self._is_linearized:
self.cache.cache_jacobian(cached_inputs, in_names, self.jac)
return self.local_data
def __update_local_data_from_cache(self, input_data, out_cached, out_jac):
"""Update the local data from the cache.
:param input_data: dict of inputs
:param out_cached: outputs retrieved from the cache
:param out_jac: jacobian retreived from the cache
"""
self.local_data = {}
self.local_data.update(input_data)
self.local_data.update(out_cached)
if out_jac is not None:
self.jac = out_jac
self._is_linearized = True
else: # Erase jacobian which is unknown
self.jac = None
self._is_linearized = False
self.check_output_data()
self._cache_was_loaded = True
def __increment_n_calls(self):
"""Increment by 1 the number of calls."""
with self._n_calls.get_lock():
self._n_calls.value += 1
def __increment_n_calls_lin(self):
"""Increment by 1 the number of calls."""
with self._n_calls_linearize.get_lock():
self._n_calls_linearize.value += 1
def __increment_exec_time(self, t_0):
"""Increment the execution time."""
curr_t = timer()
with self._exec_time.get_lock():
self._exec_time.value += curr_t - t_0
def _retreive_diff_inouts(self, force_all=False):
"""Get the list of outputs to be differentiated wrt inputs.
Get the list of outputs to be differentiated, depending on the
self._differentiated_inputs and self._differentiated_inputs attributes,
and the force_all option
"""
if force_all:
inputs = self.get_input_data_names()
outputs = self.get_output_data_names()
else:
inputs = self._differentiated_inputs
outputs = self._differentiated_outputs
return inputs, outputs
[docs] def linearize(self, input_data=None, force_all=False, force_no_exec=False):
"""Execute the linearized version of the code.
:param input_data: the input data dict needed to execute the
disciplines according to the discipline input grammar
:param force_all: if False, self._differentiated_inputs and
self.differentiated_output are used to filter
the differentiated variables
otherwise, all outputs are differentiated wrt all
inputs (Default value = False)
:param force_no_exec: if True, the discipline is not
re executed, cache is loaded anyway
"""
# TODO: remove the execution when no option exec_before_lin
# is set to True
inputs, outputs = self._retreive_diff_inouts(force_all)
if not outputs:
self.jac = {}
return self.jac
# Save inputs dict for caching
input_data = self._filter_inputs(input_data)
# if force_no_exec, we do not re execute the discipline
# otherwise, we ensure that the discipline was executed
# with the right input_data.
# This may trigger caching (see self.execute()
if not force_no_exec:
self.reset_statuses_for_run()
self.exec_for_lin = True
self.execute(input_data)
self.exec_for_lin = False
# The local_data shall be reset to their original values in case
# an input is also an output, if we don't want to keep the computed
# state (as in MDAs)
if not self._linearize_on_last_state:
self.local_data.update(input_data)
# If the caching was triggered, check if the jacobian
# was loaded,
# Or the discipline._run method also linearizes the discipline
if self._cache_was_loaded or self._is_linearized:
if self.jac:
# for cases when linearization is called
# twice with different i/o
# while cache_was_loaded=True, the check_jacobian_shape raises
# a KeyError.
try:
self._check_jacobian_shape(inputs, outputs)
return self.jac
except KeyError:
# in this case, another computation of jacobian is
# triggered.
pass
t_0 = timer()
approximate_jac = self.linearization_mode in self.APPROX_MODES
if approximate_jac:
self.jac = self._jac_approx.compute_approx_jac(outputs, inputs)
else:
self._compute_jacobian(inputs, outputs)
self.__increment_n_calls_lin()
if not approximate_jac: # Time already counted in execute()
self.__increment_exec_time(t_0)
self._check_jacobian_shape(inputs, outputs)
# Cache the Jacobian matrix
self.cache.cache_jacobian(input_data, self.get_input_data_names(), self.jac)
return self.jac
[docs] def set_jacobian_approximation(
self,
jac_approx_type=FINITE_DIFFERENCES,
jax_approx_step=1e-7,
jac_approx_n_processes=1,
jac_approx_use_threading=False,
jac_approx_wait_time=0,
):
"""Set the jacobian approximation method.
Sets the linearization mode to approx_method, sets the parameters of
the approximation for further use when calling self.linearize
:param jac_approx_type: "complex_step" or "finite_differences"
:param jax_approx_step: the step for finite differences or complex step
:param jac_approx_n_processes: maximum number of processors
on which to run
:param jac_approx_use_threading: if True, use Threads
instead of processes
to parallelize the execution
multiprocessing will copy (serialize) all the disciplines,
while threading will share all the memory
This is important to note if you want to execute the same
discipline multiple times, you shall use multiprocessing
:param jac_approx_wait_time: time waited between two forks of the
process /Thread
"""
appx = DisciplineJacApprox(
self,
approx_method=jac_approx_type,
step=jax_approx_step,
parallel=jac_approx_n_processes > 1,
n_processes=jac_approx_n_processes,
use_threading=jac_approx_use_threading,
wait_time_between_fork=jac_approx_wait_time,
)
self._jac_approx = appx
self.linearization_mode = jac_approx_type
[docs] def set_optimal_fd_step(
self,
outputs=None,
inputs=None,
force_all=False,
print_errors=False,
numerical_error=EPSILON,
):
"""Compute the optimal finite-difference step.
Compute the optimal step for a forward first order finite differences
gradient approximation.
Requires a first evaluation of perturbed functions values.
The optimal step is reached when the truncation error
(cut in the Taylor development),
and the numerical cancellation errors
(roundoff when doing f(x+step)-f(x)) are approximately equal.
Warning: this calls the discipline execution two times
per input variables.
See:
https://en.wikipedia.org/wiki/Numerical_differentiation
and
"Numerical Algorithms and Digital Representation", Knut Morken ,
Chapter 11, "Numerical Differenciation"
:param inputs: inputs wrt the linearization is made.
If None, use differentiated inputs
:param outputs: outputs of the linearization is made.
If None, use differentiated outputs
:param force_all: if True, all inputs and outputs are used
:param print_errors: if True, displays the estimated errors
:param numerical_error: numerical error associated to the calculation
of f. By default Machine epsilon (appx 1e-16),
but can be higher when
the calculation of f requires a numerical resolution
:returns: the estimated errors of truncation and cancelation error.
"""
if self._jac_approx is None:
raise ValueError(
"set_jacobian_approximation must be called "
+ "before setting an optimal step"
)
inpts, outps = self._retreive_diff_inouts(force_all=force_all)
if outputs is None or force_all:
outputs = outps
if inputs is None or force_all:
inputs = inpts
errors, steps = self._jac_approx.auto_set_step(
outputs, inputs, print_errors, numerical_error=numerical_error
)
return errors, steps
@staticmethod
def __get_len(container):
"""Return a measure of the length of an container."""
if container is None:
return -1
try:
return len(container)
except TypeError:
return 1
def _check_jacobian_shape(self, inputs, outputs):
"""Check that the jacobian is a dict of dict of 2d numpy arrays.
:param inputs: derive outputs wrt inputs
:param outputs: outputs to be derived
"""
if self.jac is None:
raise ValueError("The discipline " + self.name + " was not linearized")
out_set = set(outputs)
in_set = set(inputs)
out_jac_set = set(self.jac.keys())
if not out_set.issubset(out_jac_set):
msg = "Missing outputs in Jacobian of discipline "
msg += self.name
msg += ": " + str(out_set.difference(out_jac_set))
raise KeyError(msg)
for j_o in outputs:
j_out = self.jac[j_o]
out_dv_set = set(j_out.keys())
output_vals = self.local_data.get(j_o)
n_out_j = self.__get_len(output_vals)
if not in_set.issubset(out_dv_set):
msg = "Missing inputs " + str(in_set.difference(out_dv_set))
msg += " in Jacobian of discipline "
msg += self.name + ", for output : " + str(j_o)
raise KeyError(msg)
for j_i in inputs:
input_vals = self.local_data.get(j_i)
n_in_j = self.__get_len(input_vals)
j_mat = j_out[j_i]
expected_shape = (n_out_j, n_in_j)
if -1 in expected_shape:
# At least one of the dimensions is unknown
# Dont check shape
continue
if j_mat.shape != expected_shape:
msg = "Jacobian matrix of discipline " + str(self.name)
msg += " d " + j_o + "/d " + j_i
msg += " is not of the right shape !"
msg += "\nExpected : " + str((n_out_j, n_in_j))
msg += " got : " + str(j_mat.shape)
raise ValueError(msg)
# Discard imaginary part of Jacobian
for jac_loc in self.jac.values():
for desv, jac_out in jac_loc.items():
jac_loc[desv] = jac_out.real
@property
def cache_tol(self):
"""Accessor to the cache input tolerance."""
return self.cache.tolerance
@cache_tol.setter
def cache_tol(self, cache_tol):
"""Set to the cache input tolerance.
:param cache_tol: float, tolerance for equality
of the inputs in the cache.
If norm(inpt1-input2)<=cache_tol * norm(inpt1),
the cached data for inpt1 is returned when calling
self.execute(input2)
"""
self._set_cache_tol(cache_tol)
def _set_cache_tol(self, cache_tol):
"""Set to the cache input tolerance.
To be overloaded by subclasses
:param cache_tol: float, cache tolerance
"""
self.cache.tolerance = cache_tol or 0.0
@property
def default_inputs(self):
"""Accessor to the default inputs."""
return self._default_inputs
@default_inputs.setter
def default_inputs(self, default_inputs):
"""Set the default_inputs dict.
:param default_inputs: the default_inputs of the disciplines
it must be a dict
"""
if not isinstance(default_inputs, dict):
raise TypeError(
"MDODiscipline default inputs must be a dict,"
+ " got "
+ str(type(default_inputs))
+ " instead"
)
self._default_inputs = default_inputs
def _compute_jacobian(self, inputs=None, outputs=None):
"""Compute the jacobian matrix.
To be overloaded by subclasses, actual computation of the jacobians
:param inputs: linearization should be performed with respect
to inputs list. If None, linearization should
be performed wrt all inputs (Default value = None)
:param outputs: linearization should be performed on outputs list.
If None, linearization should be performed
on all outputs (Default value = None)
"""
def _init_jacobian(
self, inputs=None, outputs=None, with_zeros=False, fill_missing_keys=False
):
"""Initialize the jacobian dict.
:param with_zeros: if True, the matrices are set to zero
otherwise, they are empty matrices (Default value = False)
:type with_zeros: logical
:param inputs: linearization should be performed with respect
to inputs list. If None, linearization should
be performed wrt all inputs (Default value = None)
:param outputs: linearization should be performed on outputs list.
If None, linearization should be performed
on all outputs (Default value = None)
:param fill_missing_keys: if True, just fill the missing items
with zeros/empty but dont override existing data
"""
if inputs is None:
inputs_names = self._differentiated_inputs
else:
inputs_names = inputs
inputs_vals = []
for diff_name in inputs_names:
inputs_vals.append(self.get_inputs_by_name(diff_name))
if outputs is None:
outputs_names = self._differentiated_outputs
else:
outputs_names = outputs
outputs_vals = []
for diff_name in outputs_names:
outputs_vals.append(self.get_outputs_by_name(diff_name))
if with_zeros:
np_method = zeros
else:
np_method = empty
if not fill_missing_keys:
# When a key is not in the default dict, ie a function is not in
# the Jacobian; return an empty defaultdict(None)
jac = defaultdict(default_dict_factory)
for out_n, out_v in zip(outputs_names, outputs_vals):
jac_loc = defaultdict(None)
jac[out_n] = jac_loc
# When a key is not in the default dict,
# ie a variable is not in the
# Jacobian; return a defaultdict(None)
for in_n, in_v in zip(inputs_names, inputs_vals):
jac_loc[in_n] = np_method((len(out_v), len(in_v)))
self.jac = jac
else:
jac = self.jac
# Only fill the missing sub jacobians
for out_n, out_v in zip(outputs_names, outputs_vals):
jac_loc = jac.get(out_n)
if jac_loc is None:
jac_loc = defaultdict(None)
jac[out_n] = jac_loc
for in_n, in_v in zip(inputs_names, inputs_vals):
sub_jac = jac_loc.get(in_n)
if sub_jac is None:
jac_loc[in_n] = np_method((len(out_v), len(in_v)))
@property
def linearization_mode(self):
"""Accessor to the linearization mode."""
return self._linearization_mode
@linearization_mode.setter
def linearization_mode(self, linearization_mode):
"""Set the linearization_mode.
:param linearization_mode: among self.LINEARIZE_MODE_LIST
"""
if linearization_mode not in self.AVAILABLE_MODES:
msg = "Linearize '" + str(linearization_mode)
msg += "'mode unknown. Must be in "
msg += str(self.AVAILABLE_MODES)
raise ValueError(msg)
self._linearization_mode = linearization_mode
if linearization_mode in self.APPROX_MODES and self._jac_approx is None:
self.set_jacobian_approximation(linearization_mode)
[docs] def check_jacobian(
self,
input_data=None,
derr_approx=FINITE_DIFFERENCES,
step=1e-7,
threshold=1e-8,
linearization_mode="auto",
inputs=None,
outputs=None,
parallel=False,
n_processes=N_CPUS,
use_threading=False,
wait_time_between_fork=0,
auto_set_step=False,
plot_result=False,
file_path="jacobian_errors.pdf",
show=False,
figsize_x=10,
figsize_y=10,
):
"""Check if the jacobian provided by the linearize() method is correct.
:param input_data: input data dict (Default value = None)
:param derr_approx: derivative approximation method: COMPLEX_STEP
(Default value = COMPLEX_STEP)
:param threshold: acceptance threshold for the jacobian error
(Default value = 1e-8)
:param linearization_mode: the mode of linearization: direct, adjoint
or automated switch depending on dimensions
of inputs and outputs (Default value = 'auto')
:param inputs: list of inputs wrt which to differentiate
(Default value = None)
:param outputs: list of outputs to differentiate (Default value = None)
:param step: the step for finite differences or complex step
:param parallel: if True, executes in parallel
:param n_processes: maximum number of processors on which to run
:param use_threading: if True, use Threads instead of processes
to parallelize the execution
multiprocessing will copy (serialize) all the disciplines,
while threading will share all the memory
This is important to note if you want to execute the same
discipline multiple times, you shall use multiprocessing
:param wait_time_between_fork: time waited between two forks of the
process /Thread
:param auto_set_step: Compute optimal step for a forward first
order finite differences gradient approximation
:param plot_result: plot the result of the validation (computed
and approximate jacobians)
:param file_path: path to the output file if plot_result is True
:param show: if True, open the figure
:param figsize_x: x size of the figure in inches
:param figsize_y: y size of the figure in inches
:returns: True if the check is accepted, False otherwise
"""
# Do not use self._jac_approx because we may want to check complex
# step approximation with the finite differences for instance
approx = DisciplineJacApprox(
self,
derr_approx,
step,
parallel,
n_processes,
use_threading,
wait_time_between_fork,
)
if inputs is None:
inputs = self.get_input_data_names()
if outputs is None:
outputs = self.get_output_data_names()
if auto_set_step:
approx.auto_set_step(outputs, inputs, print_errors=True)
# Differentiate analytically
self.add_differentiated_inputs(inputs)
self.add_differentiated_outputs(outputs)
self.linearization_mode = linearization_mode
self.reset_statuses_for_run()
# Linearize performs execute() if needed
self.linearize(input_data)
o_k = approx.check_jacobian(
self.jac,
outputs,
inputs,
self,
threshold,
plot_result=plot_result,
file_path=file_path,
show=show,
figsize_x=figsize_x,
figsize_y=figsize_y,
)
return o_k
@property
def status(self):
"""Status accessor."""
return self._status
def _check_status(self, status):
"""Check the status according to possible statuses.
Raises an exception if status is invalid
:param status: the status to check
:type status: string
"""
if status not in [
self.STATUS_PENDING,
self.STATUS_VIRTUAL,
self.STATUS_DONE,
self.STATUS_RUNNING,
self.STATUS_FAILED,
]:
raise ValueError("Unknown status: " + str(status))
[docs] def set_disciplines_statuses(self, status):
"""Set the sub disciplines statuses.
To be implemented in subclasses.
:param status: the status
"""
[docs] def is_output_existing(self, data_name):
"""Test if output named data_name is an output of the discipline.
:param data_name: the name of the output
:returns: True if data_name is in output grammar
:rtype: logical
"""
return self.output_grammar.is_data_name_existing(data_name)
[docs] def is_all_outputs_existing(self, data_names):
"""Test if all the names in data_names are outputs of the discipline.
:param data_names: the names of the outputs
:returns: True if data_names are all in output grammar
:rtype: logical
"""
return self.output_grammar.is_all_data_names_existing(data_names)
def _is_status_ok_for_run_again(self, status):
"""Checks if the discipline can be run again.
:param status: the status
"""
return status not in [self.STATUS_RUNNING]
[docs] def reset_statuses_for_run(self):
"""Sets all the statuses to PENDING"""
if not self._is_status_ok_for_run_again(self.status):
raise ValueError(
"Cannot run discipline "
+ self.name
+ " with status "
+ str(self.status)
)
self.status = self.STATUS_PENDING
@status.setter
def status(self, status):
"""Set the statuses.
:param status: the status
"""
self._check_status(status)
self._status = status
self.notify_status_observers()
[docs] def add_status_observer(self, obs):
"""Add an observer for the status
Add an observer for the status to be notified when self changes of
status.
:param obs: the observer to add
"""
if obs not in self._status_observers:
self._status_observers.append(obs)
[docs] def remove_status_observer(self, obs):
"""Remove an observer for the status.
:param obs: the observer to remove
"""
if obs in self._status_observers:
self._status_observers.remove(obs)
[docs] def notify_status_observers(self):
"""Notify all status observers that the status has changed."""
for obs in self._status_observers[:]:
obs.update_status(self)
[docs] def store_local_data(self, **kwargs):
"""Store discipline data in local data.
:param kwargs: the data as key value pairs
"""
self.local_data.update(kwargs)
[docs] def check_output_data(self, raise_exception=True):
"""Check the output data validity.
:param raise_exception: if true, an exception is raised
when data is invalid (Default value = True)
"""
try:
self.output_grammar.load_data(self.local_data, raise_exception)
except InvalidDataException:
raise InvalidDataException("Invalid output data for: " + self.name)
[docs] def get_outputs_asarray(self):
"""Accessor for the outputs as a large numpy array.
The order is the one of self.get_all_outputs()
:returns: the outputs array
:rtype: ndarray
"""
return concatenate(list(self.get_all_outputs()))
[docs] def get_outputs_by_name(self, data_names):
"""Accessor for the outputs as a list.
:param data_names: the data names list
:returns: the data list
"""
try:
return self.get_data_list_from_dict(data_names, self.local_data)
except KeyError as err:
raise ValueError(
"Discipline " + str(self.name) + " has no output named: " + str(err)
)
[docs] def get_output_data_names(self):
"""Accessor for the output names as a list.
:returns: the data names list
"""
return self.output_grammar.get_data_names()
[docs] def get_all_outputs(self):
"""Accessor for the output data as a list of values.
The order is given by self.get_output_data_names().
:returns: the data
"""
return self.get_outputs_by_name(self.get_output_data_names())
[docs] def get_output_data(self):
"""Accessor for the output data as a dict of values.
:returns: the data dict
"""
return dict(
(k, v) for k, v in self.local_data.items() if self.is_output_existing(k)
)
[docs] def serialize(self, out_file):
"""Serialize the discipline.
:param out_file: destination file for serialization
"""
with open(out_file, "wb") as outfobj:
pickler = pickle.Pickler(outfobj, protocol=2)
pickler.dump(self)
[docs] @staticmethod
def deserialize(in_file):
"""Derialize the discipline from a file.
:param in_file: input file for serialization
:returns: a discipline instance
"""
with open(in_file, "rb") as in_fobj:
pickler = pickle.Unpickler(in_fobj)
return pickler.load()
[docs] def get_attributes_to_serialize(self): # pylint: disable=R0201
"""Define the attributes to be serialized.
Shall be overloaded by disciplines
:returns: the list of attributes names
:rtype: list
"""
# pylint warning ==> method could be a function but when overriden,
# it is a function==> self is required
return [
"residual_variables",
"output_grammar",
"name",
"local_data",
"jac",
"input_grammar",
"_status",
"cache",
"n_calls",
"n_calls_linearize",
"_differentiated_inputs",
"_differentiated_outputs",
"data_processor",
"_is_linearized",
"_linearization_mode",
"_default_inputs",
"re_exec_policy",
"exec_time",
"_cache_type",
"_cache_file_path",
"_cache_factory",
"_cache_tolerance",
"_cache_hdf_node_name",
"_linearize_on_last_state",
"_cache_was_loaded",
"grammar_type",
"comp_dir",
"exec_for_lin",
"_in_data_hash_dict",
"_jac_approx",
]
def __getstate__(self):
"""Used by pickle to define what to serialize.
:returns: the dict to serialize
"""
out_d = {}
for keep_name in self.get_attributes_to_serialize():
if keep_name not in self.__dict__:
if "_" + keep_name not in self.__dict__:
msg = "Discipline " + str(self.name)
msg += " defined attribute '" + str(keep_name)
msg += "' as required for serialization, but "
msg += " it appears to be undefined !"
raise AttributeError(msg)
prop = self.__dict__["_" + keep_name]
if not isinstance(prop, Synchronized):
raise TypeError(
"Cant handle attribute "
+ str(keep_name)
+ " serialization of undefined type"
)
# Dont serialize shared memory object,
# this is meaningless, save the value instead
out_d[keep_name] = prop.value
else:
out_d[keep_name] = self.__dict__[keep_name]
if self._cache_type == self.HDF5_CACHE:
out_d.pop("cache")
return out_d
def __setstate__(self, state_dict):
"""Used by pickle to define what to deserialize.
:param state_dict: update self dict from state_dict to deserialize
"""
self._init_shared_attrs()
self._status_observers = []
out_d = self.__dict__
shared_attrs = list(out_d.keys())
for key, val in state_dict.items():
if "_" + key not in shared_attrs:
out_d[key] = val
else:
out_d["_" + key].value = val
self.__init_cache_attr()
[docs] def get_local_data_by_name(self, data_names):
"""Accessor for the local data of the discipline as a dict of values.
:param data_names: the names of the data which will be the keys of the
dictionary
:returns: the data list
"""
try:
return self.get_data_list_from_dict(data_names, self.local_data)
except KeyError as err:
raise ValueError(
"Discipline " + str(self.name) + " has no local_data named: " + str(err)
)
[docs] @staticmethod
def is_scenario():
"""Return True if self is a scenario.
:returns: True if self is a scenario
"""
return False
[docs] @staticmethod
def get_data_list_from_dict(keys, data_dict):
"""Filter the dict from a list of keys or a single key.
If keys is a string, then the method return the value associated to the
key. If keys is a list of string, then the method return a generator of
value corresponding to the keys which can be iterated.
:param keys: a sting key or a list of keys
:param data_dict: the dict to get the data from
:returns: a data or a generator of data
"""
if isinstance(keys, string_types):
return data_dict[keys]
return (data_dict[name] for name in keys)
