import numpy
import pandas
from typing import List, Set, Union, Optional
from cosapp.drivers.abstractsetofcases import AbstractSetOfCases
from cosapp.drivers.optionaldriver import OptionalDriver
from cosapp.recorders.dataframe_recorder import DataFrameRecorder
from cosapp.utils.helpers import check_arg, is_number
from cosapp.utils.find_variables import find_variables
# TODO This does not support multipoints cases
[docs]class Influence(AbstractSetOfCases):
"""
This driver evaluate the influence between inputs and outputs (floats only)
In input :
- the inwards (including in the ports)
In output :
- the inwards (including in the ports)
- the connected ports
- the outwards
Parameters
----------
name : str
Name of the driver
owner : System, optional
:py:class:`~cosapp.systems.system.System` to which driver belongs; defaults to `None`
**kwargs : Any
Keyword arguments will be used to set driver options
"""
__slots__ = (
'input_vars', 'response_vars', 'influence_matrix', 'delta',
'influence_min_threshold', 'reference', 'found_input_vars', 'found_response_vars',
)
def __init__(
self,
name: str,
owner: Optional["cosapp.systems.System"] = None,
**kwargs
) -> None:
"""Initialize driver
Parameters
----------
name: str, optional
Name of the `Driver`.
owner: System, optional
:py:class:`~cosapp.systems.system.System` to which this driver belong; defaults to `None`.
**kwargs:
Additional keywords arguments forwarded to base class.
"""
super().__init__(name, owner, **kwargs)
self.input_vars = ["*"] # type: List[str]
# desc="Variable names varying between cases"
self.response_vars = ["*"] # type: List[str]
# desc="Variable names to monitor between cases"
self.influence_matrix = None # type: pandas.DataFrame
# desc="Influence matrix d/d output(d/d input)"
self.delta = 1e-3 # type: float
# desc="Relative influence to apply for influence matrix computation"
self.influence_min_threshold = 1e-5 # type: float
# desc="Minimum value of influence to consider"
self.reference = pandas.DataFrame() # type: pandas.DataFrame
self.found_input_vars = list() # type: List[str]
self.found_response_vars = list() # type: List[str]
[docs] def reset_response_vars(self):
"""Reset response_vars variable to its default value ['*']
"""
self.response_vars = ["*"]
[docs] def add_response_vars(self, names: Union[str, Set[str], List[str]]):
"""Add variable to be monitored.
Parameters
----------
names : Union[str, Set[str], List[str]]
List of variables to be monitored
"""
self._add_vars(names, is_input=False)
def _add_vars(self, names: Union[str, Set[str], List[str]], is_input: bool = True):
"""Add variable to be influenced or monitored.
Parameters
----------
names : Union[str, Set[str], List[str]]
List of variables to be influenced or monitored
is_input : bool
Defines if the variables is an input or not
"""
check_arg(names, 'names', (str, set, list))
if is_input and self.input_vars and self.input_vars[0] == "*":
self.input_vars.pop(0)
if not is_input and self.response_vars and self.response_vars[0] == "*":
self.response_vars.pop(0)
if isinstance(names, str):
if is_input:
self.input_vars.append(names)
self.response_vars.append(names)
else:
for name in names:
if isinstance(name, str):
if is_input:
self.input_vars.append(name)
self.response_vars.append(name)
else:
raise TypeError(
f"string expected; got {type(name).__qualname__}."
)
def _build_cases(self) -> None:
"""Build the list of cases to run during execution
"""
self.add_recorder(
DataFrameRecorder(
includes=self.input_vars + self.response_vars,
raw_output=True,
numerical_only=True,
)
)
self.found_input_vars = find_variables(
self.owner,
includes=self.input_vars,
excludes=[],
advanced_filter=lambda x: is_number(x),
outputs=False,
)
self.found_response_vars = find_variables(
self.owner,
includes=self.response_vars,
excludes=[],
advanced_filter=lambda x: is_number(x),
inputs=False,
)
# TODO('Filter on variable types to avoid bugs with strings etc')
# TODO('Support sequences')
def f(ref, idx) -> numpy.array:
case = ref.copy()
case[idx] *= 1 + self.delta
return case
case_ref = [self.owner[var] for var in self.found_input_vars]
self.cases = [case_ref]
self.cases.extend(
[f(case_ref, index) for index in range(len(self.found_input_vars))]
)
def _precase(self, case_idx, case):
"""Hook to be called before running each case."""
# TODO Use MonteCarlo machinery
for variable, value in zip(self.found_input_vars, case):
self.owner[variable] = value
def _precompute(self):
"""Set execution order, build cases and run reference case."""
super()._precompute()
self._run_reference()
OptionalDriver.set_inhibited(True)
def _run_reference(self):
"""Run the reference case (i.e. with no influence), store the result and clean the recorder."""
case = self.cases[0]
if case:
self._precase(0, case)
self.run_children()
self._postcase(0, case)
self.cases.pop(0)
data = self.recorder.export_data()
self.reference = data.iloc[
:, len(DataFrameRecorder.SPECIALS) :
] # Skip the information
self.recorder.start()
def _postcompute(self):
"""Calculate the influence matrix"""
OptionalDriver.set_inhibited(False)
super()._postcompute()
data = self.recorder.export_data()
results = data.iloc[
:, len(DataFrameRecorder.SPECIALS) :
] # Skip the information
import warnings
with warnings.catch_warnings():
warnings.simplefilter("ignore")
deltas = pandas.DataFrame(
results.values / self.reference.values - 1.0,
columns = self.reference.columns,
)
# TODO Improve coverage of no influence cases and absolute vs relative influence or mix of influences
deltas[numpy.abs(deltas) == numpy.inf] = 0
deltas[numpy.abs(results.values) < 1e-4] = 0
inputs = deltas[self.found_input_vars]
inputs_diag = inputs.values.diagonal().reshape((-1, 1))
outputs = deltas[self.found_response_vars]
inputs_diag = numpy.where(inputs_diag == 0., 1e-15, inputs_diag)
# find inputs modified during the children drivers exec (connected port, variable iterated, etc.)
coeff = numpy.where(abs(inputs_diag - self.delta) < 1e-10, 1. / inputs_diag, 0.)
self.influence_matrix = pandas.DataFrame(
outputs.values * coeff,
columns = outputs.columns,
index = inputs.columns,
).fillna(0)
[docs] def show_influence_matrix(self,
cleaned: bool = True,
styling: bool = False,
transpose: bool = False,
sort_by: str = None,
):
"""Return the influence matrix with cleaning and styling options.
Parameters
----------
cleaned : bool, optional
Defines if the returned influence matrix will be cleaned or not
styling : bool, optional
Defines if the returned influence matrix will be styled or not
transpose : bool, optional
Defines if the matrix will transposed or not before return
sort_by: str, optional
If given, will sort
"""
matrix = self.influence_matrix
if cleaned:
matrix = matrix.where((abs(matrix) > self.influence_min_threshold), 0.0)
matrix = matrix[(matrix.T != 0).any()]
matrix = matrix.loc[:, (matrix != 0).any()]
if sort_by:
matrix = matrix.reindex(
matrix[sort_by].abs().sort_values(ascending=False).index
)
if transpose:
matrix = matrix.T
if styling:
matrix = matrix.style.bar(align="mid", color=["#d65f5f", "#5fba7d"])
return matrix
