def __init__(self, Counts, L, GroupLength, prior_offset, Nyes, Ntrials, ModelResponse, value=None):
"""
Counts - nonterminal -> #h x #rules counts
L - group -> #h x 1 array
GroupLength - #groups (vector) - contains the number of trials per group
Nyes - #item ( #item = sum(GroupLength))
Ntrials - #item
ModelResponse - #h x #item - each hypothesis' response to the i'th item (1 or 0)
"""
assert sum(GroupLength) == len(Nyes) == len(Ntrials)
L = numpy.array(L)
self_update(self,locals())
self.N_groups = len(GroupLength)
self.nts = Counts.keys() # all nonterminals
self.nrules = { nt: Counts[nt].shape[1] for nt in self.nts} # number of rules for each nonterminal
self.N_hyps = Counts[self.nts[0]].shape[0]
if value is None:
value = {
'rulep': { nt: DirichletDistribution(alpha=np.ones(self.nrules[nt]), proposal_scale=1000.) for nt in self.nts },
'alpha': BetaDistribution(1,1),
'beta': BetaDistribution(1,1),
'likelihood_temperature': GammaDistribution(a=1, scale=1, proposal_scale=10.),
'prior_temperature': GammaDistribution(a=1, scale=1, proposal_scale=10.)
}
Hypothesis.__init__(self, value=value) # sets the value
def __init__(self, Counts, L, GroupLength, prior_offset, Nyes, Ntrials, ModelResponse, value=None):
"""
Counts - nonterminal -> #h x #rules counts
Hypotheses - #h
L - group -> #h x 1 array
GroupLength - #groups (vector) - contains the number of trials per group
Nyes - #item ( #item = sum(GroupLength))
Ntrials - #item
ModelResponse - #h x #item - each hypothesis' response to the i'th item (1 or 0)
"""
assert sum(GroupLength) == len(Nyes) == len(Ntrials)
L = numpy.array(L)
self_update(self,locals())
self.N_groups = len(GroupLength)
self.nts = Counts.keys() # all nonterminals
self.nrules = { nt: Counts[nt].shape[1] for nt in self.nts} # number of rules for each nonterminal
self.N_hyps = Counts[self.nts[0]].shape[0]
if value is None:
value = { nt: GibbsDirchlet(alpha=np.ones(self.nrules[nt]), proposal_scale=1000.) for nt in self.nts }
Hypothesis.__init__(self, value=value) # sets the value
def __init__(self, grammar):
"""
This takes a grammar and a regex to match variable names
"""
self_update(self, locals())
LOTProposer.__init__(self, grammar)
# check that we used "apply_" instead of "apply"
for r in self.grammar:
assert r.name is not "apply", "*** Need to use 'apply_' instead of 'apply' "
assert r.name is not "lambda_", "*** Need to use 'lambda' instead of 'lambda' "
# the asymmetry here is disturbing, but lambda is a keyword and apply is a function
self.insertable_rules = defaultdict(
list
) # Hash each nonterminal to (a,l) where a and l are the apply and lambda rules you need
for nt in self.grammar.rules.keys():
for a in filter(lambda r: (r.name == "apply_") and (r.nt == nt),
self.grammar):
for l in filter(
lambda r: isinstance(r, BVAddGrammarRule) and
(r.nt == a.to[0]) and (r.bv_args is None) and
(r.bv_type == a.to[1]), self.grammar
): # For each lambda whose "below" is the right type. bv_args are not implemented yet
self.insertable_rules[nt].append((a, l))
def __init__(self, utterance, context, possible_utterances):
"""Creates a new Utterance.
Arguments:
utterance (doc?): the word that's spoken
context (doc?): the environmental/linguistic context in which the word is spoken
possible_utterances (doc?): a set of other words we could have spoken, given the context
"""
self_update(self, locals())
def __init__(self, utterance, context, possible_utterances):
"""Creates a new Utterance.
Arguments:
utterance (doc?): the word that's spoken
context (doc?): the environmental/linguistic context in which the word is spoken
possible_utterances (doc?): a set of other words we could have spoken, given the context
"""
self_update(self, locals())
def __init__(self, current_sample, data, steps=Infinity, proposer=None, skip=0,
prior_temperature=1.0, likelihood_temperature=1.0, acceptance_temperature=1.0, trace=False,
shortcut_likelihood=True):
self_update(self,locals())
self.was_accepted = None
if proposer is None:
self.proposer = lambda x: x.propose()
self.samples_yielded = 0
self.set_state(current_sample, compute_posterior=(current_sample is not None))
self.reset_counters()
def __init__(self, value=None, n=1, proposal=None, propose_scale=1.0, propose_n=1):
self.n = n
self.propose_n = propose_n
if value is None:
value = np.random.multivariate_normal(np.array([0.0] * n), proposal)
if proposal is None:
proposal = np.eye(n) * propose_scale
propose_mask = self.get_propose_mask()
proposal = proposal * propose_mask
self.proposal = proposal
Hypothesis.__init__(self, value=value)
self_update(self, locals())
def __init__(self, grammar, fn, recurse_up=False):
"""
This manages rules that we add and subtract in the context of grammar generation. This is a class that is somewhat
in between Grammar and GrammarRule. It manages creating, adding, and subtracting the bound variable rule via "with" clause in Grammar.
NOTE: The "rule" here is the added rule, not the "bound variable" one (that adds the rule)
NOTE: If rule is None, then nothing happens
This actually could go in FunctionNode, *except* that it needs to know the grammar, which FunctionNodes do not
"""
self_update(self, locals())
self.added_rules = [] # all of the rules we added -- may be more than one from recurse_up=True
def __init__(self, nt, name, to, p=1.0, bv_prefix=None):
p = float(p)
assert p>0.0, "*** p=0 in rule %s %s %s. What are you thinking?" %(nt,name,to)
self_update(self, locals())
assert to is None or isinstance(to, list) or isinstance(to, tuple), "*** 'to' in a GrammarRule must be a list!"
for a in None2Empty(to):
assert isinstance(a,str)
if name == '':
assert (to is None) or (len(to) == 1), \
"*** GrammarRules with empty names must have only 1 argument"
def __init__(self, grammar=None, value=None, f=None, maxnodes=25, **kwargs):
if 'args' in kwargs:
assert False, "*** Use of 'args' is deprecated. Use display='...' instead."
# Save all of our keywords
self_update(self, locals())
if value is None and grammar is not None:
value = grammar.generate()
FunctionHypothesis.__init__(self, value=value, f=f, **kwargs)
self.likelihood = 0.0
self.rules_vector = None
def __init__(self, nt, name, to, p=1.0, bv_prefix=None):
p = float(p)
assert p > 0.0, "*** p=0 in rule %s %s %s. What are you thinking?" % (
nt, name, to)
self_update(self, locals())
assert to is None or isinstance(to, list) or isinstance(
to, tuple), "*** 'to' in a GrammarRule must be a list!"
for a in None2Empty(to):
assert isinstance(a, str)
if name == '':
assert (to is None) or (len(to) == 1), \
"*** GrammarRules with empty names must have only 1 argument"
def __init__(self,
nt,
name,
to,
p=1.0,
bv_prefix="y",
bv_type=None,
bv_args=None,
bv_p=None):
p = float(p)
self_update(self, locals())
assert bv_type is not None, "Did you mean to use a GrammarRule instead of a BVGrammarRule?"
assert isinstance(
bv_type, str
), "bv_type must be a string! Make sure it's not a tuple or list."
def __init__(self, grammar):
"""
This takes a grammar and a regex to match variable names
"""
self_update(self,locals())
LOTProposer.__init__(self, grammar)
# check that we used "apply_" instead of "apply"
for r in self.grammar:
assert r.name is not "apply", "*** Need to use 'apply_' instead of 'apply' "
assert r.name is not "lambda_", "*** Need to use 'lambda' instead of 'lambda' "
# the asymmetry here is disturbing, but lambda is a keyword and apply is a function
self.insertable_rules = defaultdict(list) # Hash each nonterminal to (a,l) where a and l are the apply and lambda rules you need
for nt in self.grammar.rules.keys():
for a in filter(lambda r: (r.name=="apply_") and (r.nt == nt), self.grammar):
for l in filter( lambda r: isinstance(r, BVAddGrammarRule) and (r.nt == a.to[0]) and (r.bv_args is None) and (r.bv_type==a.to[1]), self.grammar): # For each lambda whose "below" is the right type. bv_args are not implemented yet
self.insertable_rules[nt].append( (a,l) )
def __init__(self,
value=None,
n=1,
proposal=None,
propose_scale=1.0,
propose_n=1):
self.n = n
self.propose_n = propose_n
if value is None:
value = np.random.multivariate_normal(np.array([0.0] * n),
proposal)
if proposal is None:
proposal = np.eye(n) * propose_scale
propose_mask = self.get_propose_mask()
proposal = proposal * propose_mask
self.proposal = proposal
Hypothesis.__init__(self, value=value)
self_update(self, locals())
def __init__(self,
grammar=None,
value=None,
f=None,
maxnodes=25,
**kwargs):
if 'args' in kwargs:
assert False, "*** Use of 'args' is deprecated. Use display='...' instead."
# Save all of our keywords
self_update(self, locals())
if value is None and grammar is not None:
value = grammar.generate()
FunctionHypothesis.__init__(self, value=value, f=f, **kwargs)
self.likelihood = 0.0
self.rules_vector = None
def __init__(self,
current_sample,
data,
steps=Infinity,
proposer=None,
skip=0,
prior_temperature=1.0,
likelihood_temperature=1.0,
acceptance_temperature=1.0,
trace=False,
shortcut_likelihood=True):
self_update(self, locals())
self.was_accepted = None
if proposer is None:
self.proposer = lambda x: x.propose()
self.samples_yielded = 0
self.set_state(current_sample,
compute_posterior=(current_sample is not None))
self.reset_counters()
def __init__(self,
grammar,
value=None,
f=None,
node_counts=None,
maxnodes=25,
recurse_bound=25,
display="lambda recurse_, x: %s",
**kwargs):
"""
Initializer. recurse gives the name for the recursion operation internally.
"""
assert "lambda recurse_" in display, "*** RecursiveLOTHypothesis must have 'recurse_' as first display element." # otherwise it can't eval
# save recurse symbol
self.recursive_depth_bound = recurse_bound # how deep can we recurse?
self.recursive_call_depth = 0 # how far down have we recursed?
if 'args' in kwargs:
assert False, "*** Use of 'args' is deprecated. Use display='...' instead."
# Save all of our keywords
self_update(self, locals())
grammar.update_alphas() # make sure alpha/sigma arrays are initialized
if value is None:
value, self.node_counts = grammar.generate_with_counts()
else:
self.node_counts = node_counts
self.tree_size = np.sum(self.node_counts)
FunctionHypothesis.__init__(self,
value=value,
f=f,
display=display,
**kwargs)
self.likelihood = 0.0
self.compute_prior()
self.rules_vector = None
def __init__(self,
objects,
relations,
features=None,
ego=None,
distance=None):
self_update(self, locals())
self.__dict__.pop('relations')
self.parents = defaultdict(set)
self.spouses = defaultdict(set)
self.children = defaultdict(set)
for r, x, z in relations:
if r == 'parent':
self.parents[z].add(x)
self.children[x].add(z)
elif r == 'spouse':
self.spouses[x].add(z)
else:
assert False, '\t'.join([r, x, z])
def __init__(self, min, max, period):
assert max > min
assert period > 0.
self_update(self, locals())
self.ticks = 0 # how many times have we called next?
def __init__(self, nt, name, to, p=1.0, bv_prefix="y", bv_type=None, bv_args=None, bv_p=None):
p = float(p)
self_update(self, locals())
assert bv_type is not None, "Did you mean to use a GrammarRule instead of a BVGrammarRule?"
assert isinstance(bv_type, str), "bv_type must be a string! Make sure it's not a tuple or list."
def __init__(self,proposers=[],proposer_weights=[],**kwargs):
assert len(proposers) == len(proposer_weights) , "MixtureProposer.py >> __init__: different number of proposals and weights!"
self_update(self,locals())
Proposer.__init__(self,**kwargs)
def __init__(self, parent, returntype, name, args):
self_update(self,locals())
self.added_rule = None
assert self.name is None or isinstance(self.name, str)
def __init__(self, max, scale):
self_update(self, locals())
self.ticks = 0
def __init__(self, grammar, make_h, data, steps=Infinity):
self_update(self, locals())
def __init__(self, h0, data, steps=Infinity):
self_update(self, locals())
assert isinstance(h0,
LOTHypothesis) # only implemented for LOTHypothesis
self.samples_yielded = 0
def __init__(self, word, X, Y, context):
self_update(self, locals())
def __init__(self, c, alpha):
self_update(self, locals())
self.ticks = 0
def __init__(self, parent, returntype, name, args, term_type='none'):
self_update(self, locals())
self.added_rule = None
assert self.name is None or isinstance(self.name, str)
def __init__(self, h0, data, steps=Infinity):
self_update(self, locals())
assert isinstance(h0, LOTHypothesis) # only implemented for LOTHypothesis
self.samples_yielded = 0
def __init__(self, proposers=[], proposer_weights=[], **kwargs):
assert len(proposers) == len(
proposer_weights
), "MixtureProposer.py >> __init__: different number of proposals and weights!"
self_update(self, locals())
Proposer.__init__(self, **kwargs)
def __init__(self, grammar, make_h, data, steps=Infinity):
self_update(self, locals())
