LSR/fcl_symbols.py

419 lines
15 KiB
Python
Raw Normal View History

2020-06-04 19:21:01 +02:00
# -*- coding: utf-8 -*-
'''
This maps the names of IEEE and FCL options to their implementation.
For the IEEE names I'm working from the XML standard (IEEE 1855-2016).
For the FCL names I'm following fuzzylite/src/imex/FclImporter.cpp
I only record FCL names if they're different from the IEEE ones.
@author: james.power@mu.ie Created on Wed Aug 22 11:59:59 2018
'''
from collections import OrderedDict
import numpy as np
import skfuzzy.membership as fuzzmf
import skfuzzy.control.fuzzyvariable as fuzzvar
import skfuzzy.control as ctrl
import skfuzzy.control.term as fuzzterm
import extramf
import hedges
import tnorms
# ############################
# ### Membership functions ###
# ############################
# Return skfuzzy version, or one of the extras:
_IEEE_MF = { # IEEE name: (fuzz-mf, split-parameters?)
'triangular': (fuzzmf.trimf, False),
'rightlinear': (extramf.rightlinearmf, True),
'leftlinear': (extramf.leftlinearmf, True),
'pi': (fuzzmf.pimf, True),
'gaussian': (fuzzmf.gaussmf, True),
'rightgaussian': (extramf.rightgaussmf, True),
'leftgaussian': (extramf.leftgaussmf, True),
'trapezoid': (fuzzmf.trapmf, False),
's': (fuzzmf.smf, True),
'z': (fuzzmf.zmf, True),
'rectangular': (extramf.rectanglemf, True),
'singleton': (extramf.singletonmf, True),
'pointset': (extramf.pointsetmf, False),
}
# jFuzzyLogic likes these names:
_JFUZZYLOGIC_MF = {
'trian': (fuzzmf.trimf, False),
'trape': (fuzzmf.trapmf, False),
'gauss': (fuzzmf.gaussmf, True),
'gauss2': (fuzzmf.gauss2mf, True),
'gbell': (fuzzmf.gbellmf, True),
'sigm': (extramf.jfl_sigmf, True),
}
# These are some other MFs I found, mostly from fuzzylite
_FCL_MF = { # FCL name: (fuzz-mf, split-parameters?)
'bell': (extramf.fl_bellmf, True),
'concave': (extramf.concavemf, True),
'cosine': (extramf.cosinemf, True),
'gaussianproduct': (extramf.gaussprod, True),
'pishape': (fuzzmf.pimf, True),
'pointlist': (extramf.pointsetmf, False),
'ramp': (extramf.rampmf, True),
'rectangle': (extramf.rectanglemf, True),
'sigmoid': (fuzzmf.sigmf, True),
'sigmoiddifference': (fuzzmf.dsigmf, True),
'sigmoidproduct': (fuzzmf.psigmf, True),
'spike': (extramf.spikemf, True),
'sshape': (fuzzmf.smf, True),
'triangle': (fuzzmf.trimf, False),
'zshape': (fuzzmf.zmf, True),
}
# ################################
# ### Defuzzification methods: ###
# ################################
# return a string that skfuzzy.defuzzify.defuzz() can be called with.
_IEEE_DEFUZZ = {
'cog': 'centroid',
'coa': 'bisector',
'lm': 'som',
'rm': 'lom',
'mom': 'mom'
}
_FCL_DEFUZZ = {
'mm': 'mom',
'cogs': 'centroid',
# 'cogs': WeightedAverage, not implemented
# 'cogss': WeightedSum, not implemented
}
# #####################################
# ### Aggregation (AND/OR) methods: ###
# #####################################
# Note that these all return a FuzzyAggregationMethods object
# that is, you get both and/or when you lookup either one of them.
_IEEE_AND = {
'min': tnorms.MIN_MAX,
'prod': tnorms.PRODUCT_SUM,
'bdif': tnorms.BOUNDED,
'drp': tnorms.DRASTIC,
'eprod': tnorms.EINSTEIN,
'hprod': tnorms.HAMACHER,
'nilmin': tnorms.NILPOTENT,
}
_IEEE_OR = {
'max': tnorms.MIN_MAX,
'probor': tnorms.PRODUCT_SUM,
'bsum': tnorms.BOUNDED,
'drs': tnorms.DRASTIC,
'esum': tnorms.EINSTEIN,
'hsum': tnorms.HAMACHER,
'nilmax': tnorms.NILPOTENT,
}
_FCL_AND = {
'dprod': tnorms.DRASTIC,
'nmin': tnorms.NILPOTENT,
}
_FCL_OR = {
'asum': tnorms.PRODUCT_SUM, # 'algebraic sum'
'dsum': tnorms.DRASTIC,
# 'nsum' is not implemented
'nmax': tnorms.NILPOTENT,
}
_JFUZZYLOGIC_AND = {
'dmin': tnorms.DRASTIC,
'hamacher': tnorms.HAMACHER,
'nipmin': tnorms.NILPOTENT,
}
_JFUZZYLOGIC_OR = {
'asum': tnorms.PRODUCT_SUM, # 'algebraic sum'
'dmax': tnorms.DRASTIC,
'einstein': tnorms.EINSTEIN,
'nipmax': tnorms.NILPOTENT,
}
# ######################################
# ### Class to map names to objects: ###
# ######################################
class NameMapper(object):
'''
Just three dicts, mapping names to: mfs, defuzz methods and norms.
These are loaded up with the IEEE and FCL names
'''
def __init__(self):
'''
Initialise lists of known mfs, defuzz methods and and/or methods.
Can load in names from IEEE XML standard as well as FCL.
'''
self.known_mfs = {} # Membership functions
self.defuzz_methods = {} # Defuzzification methods
self.and_names = {} # And function (to be applied in rules)
self.or_names = {} # Or function (to be applied in rules)
self.hedge_names = {} # Hedge functions that can be used in rules
def load_ieee_names(self):
'''Load in the names used by the IEEE (XML) standard'''
self.known_mfs.update(_IEEE_MF)
self.defuzz_methods.update(_IEEE_DEFUZZ)
self.and_names.update(_IEEE_AND)
self.or_names.update(_IEEE_OR)
self.hedge_names.update(hedges._IEEE_HEDGES)
def load_fcl_names_too(self):
'''
Load in the names used by the IEC 1131-7 (FCL) draft standard
Note: we assume you've already loaded in the IEEE names.
'''
self.known_mfs.update(_FCL_MF)
self.defuzz_methods.update(_FCL_DEFUZZ)
self.and_names.update(_FCL_AND)
self.or_names.update(_FCL_OR)
def load_jfl_names(self):
self.known_mfs.update(_JFUZZYLOGIC_MF)
self.and_names.update(_JFUZZYLOGIC_AND)
self.or_names.update(_JFUZZYLOGIC_OR)
def _report_error(self, msg, kind, pos=None):
'''Simple error reporter (so override me)'''
assert False, '{}: {}'.format(kind, msg)
def _unsupported(self, msg):
'''Raise an 'unsupported feature' error at the current position'''
self._report_error(msg, 'unsupported feature')
def translate_mf(self, mf_name):
'''Translate a member-function name to an actual function'''
if mf_name.lower() in self.known_mfs:
return self.known_mfs[mf_name.lower()]
else:
self._unsupported('membership function "{}"'.format(mf_name))
def translate_defuzz(self, df_name):
'''Translate a given defuzz method to its skfuzzy name'''
if df_name.lower() in self.defuzz_methods:
return self.defuzz_methods[df_name.lower()]
else:
self._unsupported('defuzzify method "{}"'.format(df_name))
def translate_accu(self, accu_name):
'''
Translate a given accumulation method to its skfuzzy name.
Use skfuzzy for max/prod, otherwise select a co-norm.
'''
# First check for teh built-ins:
if accu_name.lower() == 'max':
return ctrl.accumulation_max
elif accu_name.lower() == 'prod':
return ctrl.accumulation_prod
elif accu_name.lower() in self.or_names:
return self.or_names[accu_name.lower()].or_func
else:
self._unsupported('accumulation method "{}"'.format(accu_name))
def translate_hedge(self, hedge_name):
'''
Find the named hedge function, and return the function itself.
'''
if hedge_name.lower() in self.hedge_names:
return self.hedge_names[hedge_name.lower()]
else:
self._unsupported('hedge function "{}"'.format(hedge_name))
def translate_and_or(self, and_name, or_name):
'''
Get the and/or function corresponding to the given names.
If only one specified, the other will be its dual method.
If both are specified, then take both, even if not dual.
Return a FuzzyAggregationMethods object with both functions.
'''
# First check that both names, if specified, are valid:
if and_name and and_name.lower() not in self.and_names:
self._unsupported('and method "{}"'.format(and_name))
if or_name and or_name.lower() not in self.or_names:
self._unsupported('and method "{}"'.format(or_name))
# Set up the default (is actually min/max):
fam = fuzzterm.FuzzyAggregationMethods()
# Now see if one/both have been specified
if and_name and or_name: # Set both separately:
fam.and_func = self.and_names[and_name.lower()].and_func
fam.or_func = self.or_names[or_name.lower()].or_func
elif and_name:
fam = self.and_names[and_name.lower()]
elif or_name:
fam = self.or_names[or_name.lower()]
return fam
# #######################################
# ### Symbol Table for use by parser: ###
# #######################################
class SymbolTable(object):
'''
A very simple symbol table with a list of variables and rules.
The interface mirros some methods of skfuzzy.control.ControlSystem
'''
def __init__(self, varlist=None):
'''Set up an empty symbol table; optionally supply list of variables'''
self.fb_name = None # Name of function block (if any in file)
self.variables = OrderedDict() # Map variable label to FuzzyVariable
self.all_rules = OrderedDict() # Map rule label to Rule object
self.error_on_redefine = False
if varlist:
self.add_vars(varlist)
def flag_error_on_redefine(self):
'''
Signal an error if var or rule is redefined.
Probably want to set this for files, but not for interactive use.
'''
self.error_on_redefine = True
def clear(self):
''' Empty all items in the symbol table'''
self.fb_name = None
self.variables.clear()
self.all_rules.clear()
def _report_error(self, msg, kind, pos=None):
'''Simple error reporter (so override me)'''
assert False, '{}: {}'.format(kind, msg)
def add_var(self, fvar):
'''
Add a variables to the set of those known to us.
Will potentially overwrite any previous variables with this name.
'''
assert isinstance(fvar, fuzzvar.FuzzyVariable),\
'{} should be a variable'.format(fvar)
if self.error_on_redefine and fvar.label in self.variables:
self._report_error('variable "{}"'.format(fvar.label),
'redefinition error')
self.variables[fvar.label] = fvar
def add_vars(self, varlist):
'''
Add these variables to the set of those known to us.
Will overwrite any previous variables with these names.
'''
for fvar in varlist:
self.add_var(fvar)
def get_var_defn(self, varname):
'''
Gives the variable definition for this name; error if not there.
'''
if varname not in self.variables:
self._report_error('Variable "{}" not found'.format(varname),
'scope error')
return self.variables[varname]
def is_input_var(self, varname):
'''Return true iff this varaible has been decared as input/fuzzy'''
return varname in self.variables and \
isinstance(self.variables[varname], ctrl.Antecedent)
def is_output_var(self, varname):
'''Return true iff this varaible has been decared as output/defuzzy'''
return varname in self.variables and \
isinstance(self.variables[varname], ctrl.Consequent)
def add_term_to_var(self, fvar, fterm):
if self.error_on_redefine and fterm.label in fvar.terms:
self._report_error('term "{}" of variable "{}"'
.format(fterm.label, fvar.label),
'redefinition error')
fvar[fterm.label] = fterm
@property
def antecedents(self):
"""Generator which yields Antecedents in the system."""
for node in self.variables.values():
if isinstance(node, ctrl.Antecedent):
yield node
@property
def consequents(self):
"""Generator which yields Consequents in the system."""
for node in self.variables.values():
if isinstance(node, ctrl.Consequent):
yield node
@property
def fuzzy_variables(self):
'''Return an iterator over all the variable objects'''
return self.variables.values()
@property
def rules(self):
'''Return an iterator over all the rule objects'''
return self.all_rules.values()
def add_rule(self, rule):
'''
Add this rule to the list of those known to us.
Will potentially overwrite any previous rule with the same label.
'''
assert isinstance(rule, ctrl.Rule),\
'{} should be a rule object'.format(rule)
if self.error_on_redefine and rule.label in self.all_rules:
self._report_error('rule "{}"'.format(rule.label),
'redefinition error')
self.all_rules[rule.label] = rule
return rule
def set_rule_label(self, rule, new_label):
'''
Changing the rule label has consequences for our dict,
so use this method rather than setting it directly.
Will potentially overwrite any previous rule with the same label.
'''
# Remove the old-labelled version, if there is one:
self.all_rules.pop(rule.label, None)
rule.label = new_label
self.add_rule(rule)
def __getitem__(self, key):
'''
Allows the parser to be accessed as a dict;
the key should be a variable or rule name,
returns the definition corresponding to that name (or error).
'''
if key in self.variables:
return self.variables[key]
elif key in self.all_rules:
return self.all_rules[key]
else:
self._report_error('"{}" is not a known variable or rule name'
.format(key), 'scope error')
def __str__(self):
pstr = ''
if self.fb_name:
pstr += 'Function-Block "{}"\n'.format(self.fb_name)
for var in self.fuzzy_variables:
lo, hi = np.min(var.universe), np.max(var.universe)
pstr += '{}, range := ({} .. {})\n'.format(var, lo, hi)
pstr += '{}terms: {}\n'.format(' '*12, [t for t in var.terms])
for rule in self.rules:
pstr += 'Rule {}: {}\n'.format(rule.label, rule)
return pstr