def runparts(x,datamt):
#problem: right now only recording last partition, never saving from others.
print "Start: " + str(x) + " on this many: " + str(datamt)
messup = TopN(options.top)
try:
#make new TopN for each data amount
topn= TopN(N=200, key="posterior_score")
for p in break_ctrlc(partitions):
print "Starting on partition ", p
# Now we have to go in and fill in the nodes that are nonterminals
# We can do this with generate
v = grammar.generate(deepcopy(p))
h0 = MyHypothesis(grammar, value=v)
size = datamt
data = [FunctionData(input=[],
output={'n i k': size, 'h i N': size, 'f a n': size, 'g i f': size, 'm a N': size, 'f a m': size, 'g i k': size, 'k a n': size, 'f a f': size, 'g i n': size, 'g i m': size, 'g i s': size, 's i f': size, 's i n': size, 'n i s': size, 's i m': size, 's i k': size, 'h a N': size, 'f i N': size, 'h i m': size, 'h i n': size, 'h a m': size, 'n i N': size, 'h i k': size, 'f a s': size, 'f i n': size, 'h i f': size, 'n i m': size, 'g i N': size, 'h a g': size, 's i N': size, 'n i n': size, 'f i m': size, 's i s': size, 'h i s': size, 'n a s': size, 'k a s': size, 'f i s': size, 'n i f': size, 'm i n': size, 's a s': size, 'f a g': size, 'k a g': size, 'k a f': size, 's a m': size, 'n a f': size, 'n a g': size, 'm i N': size, 's a g': size, 'f i k': size, 'k a m': size, 'n a n': size, 's a f': size, 'n a m': size, 'm a s': size, 'h a f': size, 'h a s': size, 'n a N': size, 'm i s': size, 's a n': size, 's a N': size, 'm i k': size, 'f a N': size, 'm i m': size, 'm a g': size, 'm a f': size, 'f i f': size, 'k a N': size, 'h a n': size, 'm a n': size, 'm a m': size, 'm i f': size})]
for h in break_ctrlc(MHSampler(h0, data, steps=options.steps, trace=False)):
topn.add(h)
return set(topn)
except:
#if we fail, we can return a blank TopN
return messup
def generate_unique_trees(grammar, start='START', N=1000):
"""
Yield a bunch of unique trees, produced from the grammar
"""
for _ in break_ctrlc(xrange(N)):
t = grammar.generate(start)
yield t
def runTest(self):
NSAMPLES = 10000
from LOTlib.DefaultGrammars import finiteTestGrammar as grammar
from LOTlib.Hypotheses.LOTHypothesis import LOTHypothesis
class MyH(LOTHypothesis):
@attrmem('likelihood')
def compute_likelihood(self, *args, **kwargs):
return 0.0
@attrmem('prior')
def compute_prior(self):
return grammar.log_probability(self.value)
print "# Taking MHSampler for a test run"
cnt = Counter()
h0 = MyH(grammar=grammar)
for h in break_ctrlc(
MHSampler(h0, [], steps=NSAMPLES,
skip=10)): # huh the skip here seems to be important
cnt[h] += 1
trees = list(cnt.keys())
print "# Done taking MHSampler for a test run"
## TODO: When the MCMC methods get cleaned up for how many samples they return, we will assert that we got the right number here
# assert sum(cnt.values()) == NSAMPLES # Just make sure we aren't using a sampler that returns fewer samples! I'm looking at you, ParallelTempering
Z = logsumexp([grammar.log_probability(t.value) for t in trees
]) # renormalize to the trees in self.trees
obsc = [cnt[t] for t in trees]
expc = [
exp(grammar.log_probability(t.value)) * sum(obsc) for t in trees
]
# And plot here
expc, obsc, trees = zip(*sorted(zip(expc, obsc, trees), reverse=True))
import matplotlib.pyplot as plt
plt.subplot(111)
# Log here spaces things out at the high end, where we can see it!
plt.scatter(log(range(len(trees))), expc, color="red", alpha=1.)
plt.scatter(log(range(len(trees))),
obsc,
color="blue",
marker="x",
alpha=1.)
plt.savefig('finite-sampler-test.pdf')
plt.clf()
# Do chi squared test
csq, pv = chisquare(obsc, expc)
self.assertAlmostEqual(sum(obsc), sum(expc))
# And examine
for t, c, s in zip(trees, obsc, expc):
print c, s, t
print(csq, pv), sum(obsc)
self.assertGreater(pv, 0.01, msg="Sampler failed chi squared!")
def scheme_generate():
""" This generates random scheme code with cons, cdr, and car, and evaluates it on some simple list
structures.
No inference here -- just random sampling from a grammar.
"""
example_input = [
[],
[[]],
[[], []],
[[[]]]
]
## Generate some and print out unique ones
seen = set()
for i in break_ctrlc(xrange(10000)):
x = grammar.generate('START')
if x not in seen:
seen.add(x)
# make the function node version
f = LOTHypothesis(grammar, value=x, args=['x'])
print x.log_probability(), x
for ei in example_input:
print "\t", ei, " -> ", f(ei)
def test_refs(self):
"""
Test the setting of parents and rules
"""
for _ in break_ctrlc(xrange(1000)):
x = self.G.generate()
self.assertTrue(x.check_parent_refs())
def standard_sample(make_hypothesis, make_data, skip=9, show=True, N=100, save_top='top.pkl', alsoprint='None', **kwargs):
"""
Just a simplified interface for sampling, allowing printing (showing), returning the top, and saving.
This is used by many examples, and is meant to easily allow running with a variety of parameters.
NOTE: This skip is a skip *only* on printing
**kwargs get passed to sampler
"""
if LOTlib.SIG_INTERRUPTED:
return TopN() # So we don't waste time!
h0 = make_hypothesis()
data = make_data()
best_hypotheses = TopN(N=N)
f = eval(alsoprint)
for i, h in enumerate(break_ctrlc(MHSampler(h0, data, **kwargs))):
best_hypotheses.add(h)
if show and i%(skip+1) == 0:
print i, \
h.posterior_score, \
h.prior, \
h.likelihood, \
f(h) if f is not None else '', \
qq(cleanFunctionNodeString(h))
if save_top is not None:
print "# Saving top hypotheses"
with open(save_top, 'w') as f:
pickle.dump(best_hypotheses, f)
return best_hypotheses
def plot_sampler(self, opath, sampler):
"""
Plot the sampler, for cases with many zeros where chisquared won't work well
"""
cnt = Counter()
for h in break_ctrlc(sampler):
cnt[h.value] += 1
Z = logsumexp([ self.grammar.log_probability(t) for t in self.trees]) # renormalize to the trees in self.trees
obsc = [cnt[t] for t in self.trees]
expc = [exp(self.grammar.log_probability(t)-Z)*sum(obsc) for t in self.trees]
for t, c, s in zip(self.trees, obsc, expc):
print c, "\t", s, "\t", t
expc, obsc, trees = zip(*sorted(zip(expc, obsc, self.trees), reverse=True))
import matplotlib.pyplot as plt
from numpy import log
plt.subplot(111)
# Log here spaces things out at the high end, where we can see it!
plt.scatter(log(range(len(trees))), expc, color="red", alpha=1.)
plt.scatter(log(range(len(trees))), obsc, color="blue", marker="x", alpha=1.)
plt.savefig(opath)
plt.clf()
def generate_unique_trees(grammar, start='START', N=1000):
"""
Yield a bunch of unique trees, produced from the grammar
"""
for _ in break_ctrlc(xrange(N)):
t = grammar.generate(start)
yield t
def test_check_generation_probabilities(self):
"""
Test the generation probabilities
"""
for _ in break_ctrlc(xrange(1000)):
x = self.G.generate()
self.assertTrue(x.check_generation_probabilities(self.G))
def test_lp_regenerate_propose_to(self):
# import the grammar
from Grammars import lp_regenerate_propose_to_grammar
self.G = lp_regenerate_propose_to_grammar.g
# the RegenerationProposal class
rp = RegenerationProposal(self.G)
numTests = 100
# Sample 1000 trees from the grammar, and run a chi-squared test for each of them
for i in break_ctrlc(range(numTests)):
# keep track of expected and actual counts
# expected_counts = defaultdict(int) # a dictionary whose keys are trees and values are the expected number of times we should be proposing to this tree
actual_counts = defaultdict(int) # same as expected_counts, but stores the actual number of times we proposed to a given tree
tree = self.G.generate('START')
# Regenerate some number of trees at random
numTrees = 1000
for j in range(numTrees):
newtree = rp.propose_tree(tree)[0]
# trees.append(newtree)
actual_counts[newtree] += 1
# see if the frequency with which each category of trees is generated matches the
# expected counts using a chi-squared test
chisquared, p = self.get_pvalue(tree, actual_counts, numTrees)
# print chisquared, p
# if p > 0.01/1000, test passes
self.assertTrue(p > 0.01/numTests, "Trees are not being generated according to the expected log probabilities")
if i % 10 == 0 and i != 0: print i, "lp_regenerate_propose_to tests..."
print numTests, "lp_regenerate_propose_to tests..."
def run(data_amount):
print "Starting chain on %s data points" % data_amount
data = makeVariableLexiconData(eval(options.word),
options.word,
the_context,
n=data_amount,
s=options.s,
alpha=options.alpha,
verbose=True)
h0 = KinshipLexicon(words=[options.word], alpha=options.alpha)
h0.set_word(
options.word,
LOTHypothesis(grammar, value=None, display='lambda recurse_, C, X:%s'))
hyps = TopN(N=options.top_count)
mhs = MHSampler(h0,
data,
options.steps,
likelihood_temperature=options.llt,
prior_temperature=options.prior_temp)
for samples_yielded, h in break_ctrlc(enumerate(mhs)):
if samples_yielded % 1000 == 0:
print h.prior, h.likelihood, h
hyps.add(h)
return hyps
def save_hypotheses(sampler, filename='numbergame_hypotheses.p'):
hypotheses = set()
for h in break_ctrlc(sampler):
hypotheses.add(h)
f = open(filename, "wb")
pickle.dump(hypotheses, f)
return hypotheses
def save_hypotheses(sampler, filename='numbergame_hypotheses.p'):
hypotheses = set()
for h in break_ctrlc(sampler):
hypotheses.add(h)
f = open(filename, "wb")
pickle.dump(hypotheses, f)
return hypotheses
def partitionMCMC(data,partitions):
print data
topn= TopN(N=200, key="posterior_score")
for p in break_ctrlc(partitions):
print "Starting on partition ", p
# Now we have to go in and fill in the nodes that are nonterminals
v = grammar.generate(deepcopy(p))
#h0 = MyHypothesis(grammar, value=v)
h0= make_hypothesis()
print h0
for h in break_ctrlc(MHSampler(h0, data, steps=5000, skip=0)):
# Show the partition and the hypothesis
print h.posterior_score, p, h, howyoudoin(h)
topn.add(h)
return set(topn)
def run(options, ndata):
if LOTlib.SIG_INTERRUPTED: return 0, set()
language = eval(options.LANG + "()")
data = language.sample_data(LARGE_SAMPLE)
assert len(data) == 1
# renormalize the counts
for k in data[0].output.keys():
data[0].output[k] = float(data[0].output[k] * ndata) / LARGE_SAMPLE
z = sum(data[0].output.values())
if z > 0:
best_ll = sum([(p / z) * log(p / z) for p in data[0].output.values()])
else:
best_ll = 0.0
# Now add the rules to the grammar
grammar = deepcopy(base_grammar)
for t in language.terminals(): # add in the specifics
grammar.add_rule('ATOM', "'%s'" % t, None, 1.0)
# set up the hypothesis
h0 = IncrementalLexiconHypothesis(grammar=grammar,
alphabet_size=len(language.terminals()))
h0.set_word(
0,
h0.make_hypothesis(grammar=grammar)) # make the first word at random
h0.N = 1
tn = TopN(N=options.TOP_COUNT)
for outer in xrange(options.N): # how many do we add?
if LOTlib.SIG_INTERRUPTED: return 0, set()
# and re-set the posterior or else it's something weird
h0.compute_posterior(data)
# now run mcmc
for h in break_ctrlc(MHSampler(h0, data, steps=options.STEPS)):
h.best_ll = best_ll # just store this
tn.add(copy(h))
if options.TRACE:
print h.posterior_score, h.prior, h.likelihood, h.likelihood / ndata, h
v = h()
sortedv = sorted(v.items(),
key=operator.itemgetter(1),
reverse=True)
print "{" + ', '.join(["'%s':%s" % i for i in sortedv]) + "}"
# and start from where we ended
h0 = copy(h)
h0.deepen()
return ndata, tn
def run():
h0 = make_hypothesis()
data = make_data()
for x in break_ctrlc(MHSampler(h0, data, STEPS)):
print x.posterior_score, x
for di in data:
print "\t", di.input, "->", x(*di.input), " ; should be ", di.output
def run():
h0 = make_hypothesis()
data = make_data()
for x in break_ctrlc(MHSampler(h0, data, STEPS)):
print x.posterior_score, x
for di in data:
print "\t", di.input, "->", x(
*di.input), " ; should be ", di.output
def __call__(self, generator):
"""Pass this a generator, add each element as it's yielded.
This allows us to make a pipeline. See Example in main docstring: '# Or as a generator...'.
"""
if hasattr(generator, 'data'):
self.data = generator.data
for sample in break_ctrlc(generator):
self.add(sample)
yield sample
def __call__(self, generator):
"""Pass this a generator, add each element as it's yielded.
This allows us to make a pipeline. See Example in main docstring: '# Or as a generator...'.
"""
if hasattr(generator, 'data'):
self.data = generator.data
for sample in break_ctrlc(generator):
self.add(sample)
yield sample
def test_eq(self):
counter = 0
for i in break_ctrlc(xrange(10000)):
x = self.G.generate()
y = self.G.generate()
if pystring(x) == pystring(y):
counter += 1
# print(counter)
#print( pystring(x)+'\n'+ pystring(y)+'\n')
self.assertEqual( pystring(x) == pystring(y), x == y, "Without bvs, the pystrings should be the same")
def run(options, ndata):
if LOTlib.SIG_INTERRUPTED: return 0, set()
language = eval(options.LANG+"()")
data = language.sample_data(LARGE_SAMPLE)
assert len(data) == 1
# renormalize the counts
for k in data[0].output.keys():
data[0].output[k] = float(data[0].output[k] * ndata) / LARGE_SAMPLE
z = sum(data[0].output.values())
if z > 0:
best_ll = sum([ (p/z)*log(p/z) for p in data[0].output.values() ])
else:
best_ll = 0.0
# Now add the rules to the grammar
grammar = deepcopy(base_grammar)
for t in language.terminals(): # add in the specifics
grammar.add_rule('ATOM', "'%s'" % t, None, 1.0)
# set up the hypothesis
h0 = IncrementalLexiconHypothesis(grammar=grammar, alphabet_size=len(language.terminals()))
h0.set_word(0, h0.make_hypothesis(grammar=grammar)) # make the first word at random
h0.N = 1
tn = TopN(N=options.TOP_COUNT)
for outer in xrange(options.N): # how many do we add?
if LOTlib.SIG_INTERRUPTED: return 0, set()
# and re-set the posterior or else it's something weird
h0.compute_posterior(data)
# now run mcmc
for h in break_ctrlc(MHSampler(h0, data, steps=options.STEPS)):
h.best_ll = best_ll # just store this
tn.add(copy(h))
if options.TRACE:
print h.posterior_score, h.prior, h.likelihood, h.likelihood / ndata, h
v = h()
sortedv = sorted(v.items(), key=operator.itemgetter(1), reverse=True )
print "{" + ', '.join(["'%s':%s"% i for i in sortedv]) + "}"
# and start from where we ended
h0 = copy(h)
h0.deepen()
return ndata, tn
def runTest(self):
NSAMPLES = 10000
from LOTlib.DefaultGrammars import finiteTestGrammar as grammar
from LOTlib.Hypotheses.LOTHypothesis import LOTHypothesis
class MyH(LOTHypothesis):
@attrmem('likelihood')
def compute_likelihood(self, *args, **kwargs):
return 0.0
@attrmem('prior')
def compute_prior(self):
return grammar.log_probability(self.value)
print "# Taking MHSampler for a test run"
cnt = Counter()
h0 = MyH(grammar=grammar)
for h in break_ctrlc(MHSampler(h0, [], steps=NSAMPLES, skip=10)): # huh the skip here seems to be important
cnt[h] += 1
trees = list(cnt.keys())
print "# Done taking MHSampler for a test run"
## TODO: When the MCMC methods get cleaned up for how many samples they return, we will assert that we got the right number here
# assert sum(cnt.values()) == NSAMPLES # Just make sure we aren't using a sampler that returns fewer samples! I'm looking at you, ParallelTempering
Z = logsumexp([grammar.log_probability(t.value) for t in trees]) # renormalize to the trees in self.trees
obsc = [cnt[t] for t in trees]
expc = [exp( grammar.log_probability(t.value))*sum(obsc) for t in trees]
# And plot here
expc, obsc, trees = zip(*sorted(zip(expc, obsc, trees), reverse=True))
import matplotlib.pyplot as plt
plt.subplot(111)
# Log here spaces things out at the high end, where we can see it!
plt.scatter(log(range(len(trees))), expc, color="red", alpha=1.)
plt.scatter(log(range(len(trees))), obsc, color="blue", marker="x", alpha=1.)
plt.savefig('finite-sampler-test.pdf')
plt.clf()
# Do chi squared test
csq, pv = chisquare(obsc, expc)
self.assertAlmostEqual(sum(obsc), sum(expc))
# And examine
for t, c, s in zip(trees, obsc, expc):
print c, s, t
print (csq, pv), sum(obsc)
self.assertGreater(pv, 0.01, msg="Sampler failed chi squared!")
def generate_data(data_size):
all_words = target.all_words()
data = []
for i in break_ctrlc(xrange(data_size)):
# a context is a set of men, pirates, and everything. functions are applied to this to get truth values
context = sample_context()
word = target.sample_utterance(all_words, context)
data.append(
UtteranceData(utterance=word,
context=context,
possible_utterances=all_words))
return data
def runme(chain, dataamt):
if LOTlib.SIG_INTERRUPTED: return ()
data = make_data(dataamt)
tn = TopN(options.top)
h0 = make_hypothesis()
for h in break_ctrlc(MHSampler(h0, data, steps=options.steps, skip=0)):
# print h.posterior_score, h.prior, h.likelihood, h
h.likelihood_per_data = h.likelihood/dataamt
tn.add(h)
return tn
def myrun(observed_set):
if LOTlib.SIG_INTERRUPTED:
return set()
h0 = NumberGameHypothesis(grammar=grammar)
data = [FunctionData(input=[], output=observed_set, alpha=ALPHA)]
tn = TopN(N=options.TOP_COUNT)
for h in break_ctrlc(MHSampler(h0, data, steps=options.STEPS)):
tn.add(h)
print "# Finished %s" % str(observed_set)
return set(tn.get_all())
def myrun(observed_set):
if LOTlib.SIG_INTERRUPTED:
return set()
h0 = NumberGameHypothesis(grammar=grammar)
data = [FunctionData(input=[], output=observed_set, alpha=ALPHA)]
tn = TopN(N=options.TOP_COUNT)
for h in break_ctrlc(MHSampler(h0, data, steps=options.STEPS)):
tn.add(h)
print "# Finished %s" % str(observed_set)
return set(tn.get_all())
def run(data_amount):
print "Starting chain on %s data points"%data_amount
data = makeLexiconData(target, four_gen_tree_context, n=data_amount, alpha=options.alpha)
h0 = KinshipLexicon(alpha=options.alpha)
for w in target_words:
h0.set_word(w, LOTHypothesis(my_grammar, args=['recurse_','C', 'X']))
hyps = TopN(N=options.top_count)
mhs = MHSampler(h0, data, options.steps, likelihood_temperature=options.llt, prior_temperature=options.prior_temp)
for samples_yielded, h in break_ctrlc(enumerate(mhs)):
hyps.add(h)
return hyps
def test_setto(self):
"""
Test the operation of setting a function node to another.
"""
for _ in break_ctrlc(xrange(1000)):
x = self.G.generate()
x0 = copy(x)
y = self.G.generate()
y_subnodes = y.subnodes()
x.setto(y)
self.assertTrue(x.check_parent_refs())
self.assertTrue(x.check_generation_probabilities(self.G))
for xi in x:
self.assertTrue(xi in y_subnodes)
def standard_sample(make_hypothesis, make_data, show_skip=9, show=True, N=100, save_top='top.pkl', alsoprint='None', **kwargs):
"""
Just a simplified interface for sampling, allowing printing (showing), returning the top, and saving.
This is used by many examples, and is meant to easily allow running with a variety of parameters.
NOTE: This skip is a skip *only* on printing
**kwargs get passed to sampler
"""
if LOTlib.SIG_INTERRUPTED:
return TopN() # So we don't waste time!
h0 = make_hypothesis()
data = make_data()
best_hypotheses = TopN(N=N)
f = eval(alsoprint)
sampler = MHSampler(h0, data, **kwargs)
# # TODO change acceptance temperature over times
# sampler.acceptance_temperature = 0.5
for i, h in enumerate(break_ctrlc(sampler)):
# if i % 10000 == 0 and i != 0:
# sampler.acceptance_temperature = min(1.0, sampler.acceptance_temperature+0.1)
# print '='*50
# print 'change acc temperature to', sampler.acceptance_temperature
best_hypotheses.add(h)
if show and i%(show_skip+1) == 0:
print i, \
h.posterior_score, \
h.prior, \
h.likelihood, \
f(h) if f is not None else '', \
qq(cleanFunctionNodeString(h))
if save_top is not None:
print "# Saving top hypotheses"
with open(save_top, 'w') as f:
pickle.dump(best_hypotheses, f)
return best_hypotheses
def run(options, ndata):
"""
This out on the DATA_RANGE amounts of data and returns all hypotheses in top count
"""
if LOTlib.SIG_INTERRUPTED:
return set()
language = eval(options.LANG + "()")
data = language.sample_data(LARGE_SAMPLE)
assert len(data) == 1
# renormalize the counts
for k in data[0].output.keys():
data[0].output[k] = float(data[0].output[k] * ndata) / LARGE_SAMPLE
# print data
# Now add the rules to the grammar
grammar = deepcopy(base_grammar)
for t in language.terminals(): # add in the specifics
grammar.add_rule("ATOM", q(t), None, 2)
h0 = AugustHypothesis(grammar=grammar, display="lambda recurse_ :%s")
print "# Starting on ", h0
tn = TopN(N=options.TOP_COUNT)
# print h0.compute_posterior(data)
# for i, h in enumerate(break_ctrlc(MHSampler(h0, data, steps=options.STEPS))):
# # for h in MHSampler(h0, data, steps=options.STEPS, trace=True):
# print h.posterior_score, h
# print getattr(h, 'll_counts', None)
with open(
prefix + "hypotheses_" + options.LANG + "_" + str(rank) + "_" + str(ndata) + "_" + suffix + ".txt", "a"
) as ofile:
for i, h in enumerate(break_ctrlc(MHSampler(h0, data, steps=options.STEPS))):
tn.add(h)
# print h.posterior_score, getattr(h, 'll_counts', None), h
if i % options.SKIP == 0 and h.posterior_score > -Infinity:
print >> ofile, i, ndata, h.posterior_score, h.prior, h.likelihood, h.likelihood / ndata
print >> ofile, getattr(h, "ll_counts", None)
print >> ofile, h, "\0" # must add \0 when not Lexicon
return tn
def run(options, ndata):
"""
This out on the DATA_RANGE amounts of data and returns all hypotheses in top count
"""
if LOTlib.SIG_INTERRUPTED:
return 0, set()
language = eval(options.LANG+"()")
data = language.sample_data(LARGE_SAMPLE)
assert len(data) == 1
# renormalize the counts
for k in data[0].output.keys():
data[0].output[k] = float(data[0].output[k] * ndata) / LARGE_SAMPLE
#print data
# Now add the rules to the grammar
grammar = deepcopy(base_grammar)
for t in language.terminals(): # add in the specifics
grammar.add_rule('ATOM', q(t), None, 2)
h0 = IncrementalLexiconHypothesis(grammar=grammar)
tn = TopN(N=options.TOP_COUNT)
for outer in xrange(options.N): # how many do we add?
# add to the grammar
grammar.add_rule('SELFF', '%s' % (outer), None, 1.0)
# Add one more to the number of words here
h0.set_word(outer, h0.make_hypothesis(grammar=grammar))
h0.N = outer+1
assert len(h0.value.keys())==h0.N==outer+1
# now run mcmc
for h in break_ctrlc(MHSampler(h0, data, steps=options.STEPS)):
tn.add(h)
print h.posterior_score, h
print getattr(h, 'll_counts', None)
# and start from where we ended
h0 = deepcopy(h) # must deepcopy
return ndata, tn
def run(options, ndata):
"""
This out on the DATA_RANGE amounts of data and returns all hypotheses in top count
"""
if LOTlib.SIG_INTERRUPTED:
return 0, set()
language = eval(options.LANG+"()")
data = language.sample_data(LARGE_SAMPLE)
assert len(data) == 1
# renormalize the counts
for k in data[0].output.keys():
data[0].output[k] = float(data[0].output[k] * ndata) / LARGE_SAMPLE
#print data
# Now add the rules to the grammar
grammar = deepcopy(base_grammar)
for t in language.terminals(): # add in the specifics
grammar.add_rule('ATOM', q(t), None, 2)
h0 = IncrementalLexiconHypothesis(grammar=grammar)
tn = TopN(N=options.TOP_COUNT)
for outer in xrange(options.N): # how many do we add?
# add to the grammar
grammar.add_rule('SELFF', '%s' % (outer), None, 1.0)
# Add one more to the number of words here
h0.set_word(outer, h0.make_hypothesis(grammar=grammar))
h0.N = outer+1
assert len(h0.value.keys())==h0.N==outer+1
# now run mcmc
for h in break_ctrlc(MHSampler(h0, data, steps=options.STEPS)):
tn.add(h)
# print h.posterior_score, h
# print getattr(h, 'll_counts', None)
# and start from where we ended
h0 = deepcopy(h) # must deepcopy
return ndata, tn
def run(data_pts):
print "Start run on ", str(data_pts)
y = [pt.Y for pt in data_pts]
filename = "".join(y)
hyps = TopN(N=options.TOP_COUNT)
h0 = KinshipLexicon(alpha=options.ALPHA)
h0.set_word('Word', LOTHypothesis(my_grammar, value=None, display='lambda recurse_, C, X:%s'))
mhs = MHSampler(h0, data_pts, options.STEPS, likelihood_temperature=options.llt)
for samples_yielded, h in break_ctrlc(enumerate(mhs)):
hyps.add(h)
with open(options.OUT_PATH + filename + '.pkl', 'w') as f:
pickle.dump(hyps, f)
return filename, hyps
def run():
"""A version that cares more about recent data, showing how to use
Hypotheses.DecayedLikelihoodHypothesis.
"""
G = grammar
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Create an initial hypothesis
# This is where we set a number of relevant variables -- whether to use RR, alpha, etc.Z
h0 = MyHypothesis(G, ll_decay=1.0, rrAlpha=1.0, args=['x'])
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Run the MH
# Run the vanilla sampler. Without steps, it will run infinitely
# this prints out posterior (posterior_score), prior, likelihood,
for h in break_ctrlc(MHSampler(h0, data, 10000, skip=100)):
print h.posterior_score, h.prior, h.likelihood, q(h)
def run(options, ndata):
"""
This out on the DATA_RANGE amounts of data and returns all hypotheses in top count
"""
if LOTlib.SIG_INTERRUPTED:
return set()
language = eval(options.LANG+"()")
data = language.sample_data(LARGE_SAMPLE)
assert len(data) == 1
# renormalize the counts
for k in data[0].output.keys():
data[0].output[k] = float(data[0].output[k] * ndata) / LARGE_SAMPLE
# print data
# Now add the rules to the grammar
grammar = deepcopy(base_grammar)
for t in language.terminals(): # add in the specifics
grammar.add_rule('ATOM', q(t), None, 2)
h0 = AugustHypothesis(grammar=grammar, display="lambda recurse_ :%s")
print "# Starting on ", h0
tn = TopN(N=options.TOP_COUNT)
# print h0.compute_posterior(data)
# for i, h in enumerate(break_ctrlc(MHSampler(h0, data, steps=options.STEPS))):
# # for h in MHSampler(h0, data, steps=options.STEPS, trace=True):
# print h.posterior_score, h
# print getattr(h, 'll_counts', None)
with open(prefix+'hypotheses_'+options.LANG+'_'+str(rank)+'_'+str(ndata)+'_'+suffix+".txt", 'a') as ofile:
for i, h in enumerate(break_ctrlc(MHSampler(h0, data, steps=options.STEPS))):
tn.add(h)
# print h.posterior_score, getattr(h, 'll_counts', None), h
if i%options.SKIP == 0 and h.posterior_score > -Infinity:
print >>ofile, i, ndata, h.posterior_score, h.prior, h.likelihood, h.likelihood/ndata
print >>ofile, getattr(h,'ll_counts', None)
print >>ofile, h, '\0' # must add \0 when not Lexicon
return tn
def run(data_amount):
print "Starting chain on %s data points" % data_amount
data = makeVariableLexiconData(eval(options.word), options.word, the_context, n=data_amount, s=options.s,
alpha=options.alpha, verbose=True)
h0 = KinshipLexicon(words=[options.word], alpha=options.alpha)
h0.set_word(options.word, LOTHypothesis(grammar, value=None, args=['recurse_', 'C', 'X']))
hyps = TopN(N=options.top_count)
mhs = MHSampler(h0, data, options.steps, likelihood_temperature=options.llt,
prior_temperature=options.prior_temp)
for samples_yielded, h in break_ctrlc(enumerate(mhs)):
#if samples_yielded % 100 == 0:
# print h.prior, h.likelihood, h
hyps.add(h)
return hyps
def run_mh():
"""Run the MH."""
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# somewhat weirdly, we'll make an upper node above "START" for the two concepts
# and require it to check if concept (an argument below) is 'A'
grammar.add_rule('TWO_CONCEPT_START', 'if_', ['(concept==\'A\')', 'START', 'START'], 1.0)
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Create an initial hypothesis
# This is where we set a number of relevant variables -- whether to use RR, alpha, etc.
# Here we give args as "concept" (used in TWO_CONCEPT_START above) and "x"
h0 = RationalRulesLOTHypothesis(grammar=grammar, rrAlpha=1.0, ALPHA=0.9, start='TWO_CONCEPT_START',
args=['concept', 'x'])
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Run the vanilla sampler. Without steps, it will run infinitely
# this prints out posterior (posterior_score), prior, likelihood,
for h in break_ctrlc(MHSampler(h0, data, 10000, skip=100)):
print h.posterior_score, h.prior, h.likelihood, q(h)
def run(options, ndata):
"""
This out on the DATA_RANGE amounts of data and returns all hypotheses in top count
"""
if LOTlib.SIG_INTERRUPTED:
return set()
language = eval(options.LANG + "()")
data = language.sample_data(LARGE_SAMPLE)
assert len(data) == 1
# renormalize the counts
for k in data[0].output.keys():
data[0].output[k] = float(data[0].output[k] * ndata) / LARGE_SAMPLE
print data
# Now add the rules to the grammar
grammar = deepcopy(base_grammar)
for t in language.terminals(): # add in the specifics
grammar.add_rule('ATOM', q(t), None, 2)
h0 = AugustHypothesis(grammar=grammar, display="lambda recurse_ :%s")
tn = TopN(N=options.TOP_COUNT)
for i, h in enumerate(break_ctrlc(MHSampler(h0, data,
steps=options.STEPS))):
print h.posterior_score, h
print getattr(h, 'll_counts', None)
# with open(prefix+'hypotheses_'+options.LANG+'_'+str(rank)+'_'+str(ndata)+'_'+suffix+".txt", 'a') as ofile:
#
# for i, h in enumerate(break_ctrlc(MHSampler(h0, data, steps=options.STEPS))):
# tn.add(h)
# # print h.posterior_score, getattr(h, 'll_counts', None), h
# if i%options.SKIP == 0:
# print >>ofile, "\n"
# print >>ofile, i, ndata, h.posterior_score, h.prior, h.likelihood, h.likelihood/ndata
# print >>ofile, getattr(h,'ll_counts', None),
# print >>ofile, h # ends in \0 so we can sort with sort -g -z
return tn
def run(options, ndata):
"""
This out on the DATA_RANGE amounts of data and returns all hypotheses in top count
"""
if LOTlib.SIG_INTERRUPTED:
return set()
language = eval(options.LANG + "()")
data = language.sample_data(LARGE_SAMPLE)
assert len(data) == 1
# renormalize the counts
for k in data[0].output.keys():
data[0].output[k] = float(data[0].output[k] * ndata) / LARGE_SAMPLE
print data
# Now add the rules to the grammar
grammar = deepcopy(base_grammar)
for t in language.terminals(): # add in the specifics
grammar.add_rule("ATOM", q(t), None, 2)
h0 = AugustHypothesis(grammar=grammar, display="lambda recurse_ :%s")
tn = TopN(N=options.TOP_COUNT)
for i, h in enumerate(break_ctrlc(MHSampler(h0, data, steps=options.STEPS))):
print h.posterior_score, h
print getattr(h, "ll_counts", None)
# with open(prefix+'hypotheses_'+options.LANG+'_'+str(rank)+'_'+str(ndata)+'_'+suffix+".txt", 'a') as ofile:
#
# for i, h in enumerate(break_ctrlc(MHSampler(h0, data, steps=options.STEPS))):
# tn.add(h)
# # print h.posterior_score, getattr(h, 'll_counts', None), h
# if i%options.SKIP == 0:
# print >>ofile, "\n"
# print >>ofile, i, ndata, h.posterior_score, h.prior, h.likelihood, h.likelihood/ndata
# print >>ofile, getattr(h,'ll_counts', None),
# print >>ofile, h # ends in \0 so we can sort with sort -g -z
return tn
def scheme_generate():
""" This generates random scheme code with cons, cdr, and car, and evaluates it on some simple list
structures.
No inference here -- just random sampling from a grammar.
"""
## Generate some and print out unique ones
seen = set()
for i in break_ctrlc(xrange(10000)):
x = grammar.generate('START')
if x not in seen:
seen.add(x)
# make the function node version
f = LOTHypothesis(grammar, value=x, args=['x'])
print x.log_probability(), x
for ei in example_input:
print "\t", ei, " -> ", f(ei)
def run(data_amount):
print "Starting chain on %s data points"%data_amount
data = makeLexiconData(target, four_gen_tree_context, n=data_amount, alpha=options.alpha, verbose=True)
h0 = KinshipLexicon(alpha=options.alpha)
for w in target_words:
h0.set_word(w, LOTHypothesis(my_grammar, display='lambda recurse_, C, X: %s'))
hyps = TopN(N=options.top_count)
mhs = MHSampler(h0, data, options.steps, likelihood_temperature=options.llt, prior_temperature=options.prior_temp)
for samples_yielded, h in break_ctrlc(enumerate(mhs)):
hyps.add(h)
import pickle
print 'Writing ' + data[0].X + data[0].Y + str(data_amount) + data[0].word + '.pkl'
with open('Chains/' + data[0].X + data[0].Y + str(data_amount) + data[0].word + '.pkl', 'w') as f:
pickle.dump(hyps, f)
return hyps
def test_substitute(self):
"""
Test how substitution works
"""
for _ in break_ctrlc(xrange(1000)):
x = self.G.generate()
y, _ = x.sample_subnode()
oldy = copy(y)
repl = self.G.generate(y.returntype)
# pick a novel replacemnet (must NOT equal y)
# NOTE: We can't just rejection sample here to pick repl because it may not be possible for a given x
if repl == y: continue
x.replace_subnodes(lambda z: z==y, repl)
self.assertTrue(x.check_parent_refs())
# We cannot check generation_probabilites because replace_subnodes breaks that!
# and ensure that y is not left!
for xi in x:
self.assertTrue(xi != oldy, "\n%s\n%s\n%s\n%s"%(x,y,repl,xi))
def run(hypothesis, data_amount):
print "Starting chain on %s data points" % data_amount
data = makeLexiconData(target,
four_gen_tree_context,
n=data_amount,
alpha=options.alpha,
verbose=True)
h0 = KinshipLexicon(alpha=options.alpha)
for w in target_words:
h0.set_word(
w,
LOTHypothesis(grammar=my_grammar,
value=hypothesis.value[w].value,
display='lambda recurse_, C, X: %s'))
hyps = TopN(N=options.top_count)
mhs = MHSampler(h0,
data,
options.steps,
likelihood_temperature=options.llt,
prior_temperature=options.prior_temp)
for samples_yielded, h in break_ctrlc(enumerate(mhs)):
if samples_yielded % 100 == 0:
pass #print h.likelihood, h.prior, h
hyps.add(h)
import pickle
print 'Writing ' + data[0].X + data[0].Y + str(
data_amount) + data[0].word + '.pkl'
with open(
'Chains/' + data[0].X + data[0].Y + str(data_amount) +
data[0].word + '.pkl', 'w') as f:
pickle.dump(hyps, f)
return hyps
def evaluate_sampler(self, sampler):
cnt = Counter()
for h in break_ctrlc(sampler):
cnt[h.value] += 1
## TODO: When the MCMC methods get cleaned up for how many samples they return, we will assert that we got the right number here
# assert sum(cnt.values()) == NSAMPLES # Just make sure we aren't using a sampler that returns fewer samples! I'm looking at you, ParallelTempering
Z = logsumexp([self.grammar.log_probability(t) for t in self.trees]) # renormalize to the trees in self.trees
obsc = [cnt[t] for t in self.trees]
expc = [exp( self.grammar.log_probability(t))*sum(obsc) for t in self.trees]
csq, pv = chisquare(obsc, expc)
assert abs(sum(obsc) - sum(expc)) < 0.01
# assert min(expc) > 5 # or else chisq sux
for t, c, s in zip(self.trees, obsc, expc):
print c, s, t
print (csq, pv), sum(obsc)
self.assertGreater(pv, PVALUE, msg="Sampler failed chi squared!")
return csq, pv
def evaluate_sampler(my_sampler, print_every=1000, out_aggregate=sys.stdout, trace=False, pthreshold=0.999, prefix=""):
"""
Print the stats for a single sampler run
*my_sampler* -- a generator of samples
print_every -- display the output every this many steps
out_hypothesis -- where we put hypothesis stats
out_aggregate -- where we put aggregate stats
trace -- print every sample
prefix -- display before lines
"""
visited_at = defaultdict(list)
startt = time()
for n, s in break_ctrlc(enumerate(my_sampler)): # each sample should have an .posterior_score defined
if trace: print "#", n, s
visited_at[s].append(n)
if (n%print_every)==0 and n>0:
post = sorted([x.posterior_score for x in visited_at.keys()], reverse=True) # the unnormalized posteriors of everything found
ll = sorted([x.likelihood for x in visited_at.keys()], reverse=True)
Z = logsumexp(post) # just compute total probability mass found -- the main measure
# determine how many you need to get pthreshold of the posterior mass
J=0
while J < len(post):
if logsumexp(post[J:]) < Z + log(1.0-pthreshold):
break
J += 1
out_aggregate.write('\t'.join(map(str, [prefix, n, r3(time()-startt), r5(Z), r5(post[0]), J, len(post)] )) + '\n')
out_aggregate.flush()
return
class ParticleSwarmPriorResample(ParticleSwarm):
"""
Like ParticleSwarm, but resamples from the prior
"""
def refresh(self):
"""
Resample by resampling those below the median from the prior.
"""
m = median(self.chainZ)
for i in range(self.nchains):
if self.chainZ[i] < m:
self.chains[i] = self.make_h0(**self.kwargs)
self.chainZ[i] = -Infinity # reset this
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
if __name__ == "__main__":
from LOTlib.Examples.Number.Global import generate_data, make_h0
data = generate_data(300)
ps = ParticleSwarm(make_h0, data)
for h in break_ctrlc(ps):
print h.posterior_score, h
if len(ps.seen) > 0:
print "#", sorted(ps.seen,
key=lambda x: x.posterior_score,
reverse=True)[0]
probs=[v.posterior_score for v in population],
log=True)
try:
kid = mutate(crossover(mom, dad))
except (ProposalFailedException, NodeSamplingException):
continue
kid.compute_posterior(data)
yield kid
nextpopulation.append(kid)
# # if MH_acceptance(population[i].posterior_score, kid.posterior_score, 0.0):
# if kid.posterior_score > population[i].posterior_score:
# population[i] = kid
# yield kid
population = nextpopulation
if __name__ == "__main__":
from LOTlib import break_ctrlc
from LOTlib.Examples.Number.Model import make_hypothesis, make_data
from LOTlib.Miscellaneous import qq
data = make_data(400)
for h in break_ctrlc(
genetic_algorithm(make_hypothesis, data, mutate_lot,
crossover_lot)):
print h.posterior_score, h.get_knower_pattern(), qq(h)
def __iter__(self):
for i, t in enumerate(self.grammar.enumerate()):
if i >= self.steps:
raise StopIteration
h = self.make_h(value=t)
h.compute_posterior(self.data)
yield h
if __name__ == "__main__":
from LOTlib import break_ctrlc
#from LOTlib.Examples.Number.Shared import grammar, make_h0, generate_data
#data = generate_data(100)
#from LOTlib.Examples.RegularExpression.Shared import grammar, make_h0, data
from LOTlib.Examples.Magnetism.Simple import make_data, make_hypothesis
from LOTlib.Examples.Magnetism.Simple.Grammar import grammar
#from LOTlib.Examples.RationalRules.Shared import grammar, data, make_h0
for h in break_ctrlc(
EnumerationInference(grammar,
make_hypothesis,
make_data(),
steps=10000)):
print h.posterior_score, h
def print_subtree_adaptations(grammar,
hypotheses,
posteriors,
subtrees,
relative_KL=True):
"""
Determine how useful it would be to explicitly define each subtree in H across
all of the (corresponding) posteriors, as measured by KL from prior to posterior
- hypotheses - a list of LOThypotheses
- posteriors - [ [P(h|data) for h in hypotheies] x problems ]
- subtrees - a collection of (possibly partial) subtrees to try adapting
We treat hyps as a fixed finite hypothesis space, and assume every subtree considered
is *not* derived compositionally (although this could change in future versions)
p - the probability of going to kids in randomly generating a subtree
subtree_multiplier - how many times we sample a subtree from *each* node in each hypothesis
relative_KL - compute summed KL divergence absolutely, or relative to the h.compute_prior()?
"""
# compute the normalized posteriors
Ps = map(lognormalize, posteriors)
# Compute the baseline KL divergence so we can score relative to this
if relative_KL:
oldpriors = lognormalize(
numpy.array([h.compute_prior() for h in hypotheses]))
KL0s = [sum(exp(oldpriors) * (oldpriors - P)) for P in Ps]
else:
KL0s = [
1.0 for P in Ps
] # pretend everything just had KL of 1, so we score relatively
## Now process each, starting with the most simple
for t in break_ctrlc(
sorted(subtrees,
key=lambda t: grammar.log_probability(t),
reverse=True)):
# Get some stats on t:
tlp = grammar.log_probability(t)
tnt = count_identical_nonterminals(
t.returntype, t) # How many times is this nonterminal used?
# How many matches of t are there in each H?
m = numpy.array(
[count_subtree_matches(t, h.value) for h in hypotheses])
## TODO: There is a complication: partial patterns matching themselves.
## For simplicity, we'll just take the *first* match, seetting max(m)=1
## In the future, we should change this to correctly handle and count
## partial matches matching themselves
m = (m >= 1) * 1
assert max(m) == 1, "Error: " + str(t) + "\t" + str(m)
# How many times is the nonterminal used, NOT counting in t?
nt = numpy.array([
count_identical_nonterminals(t.returntype, h.value)
for h in hypotheses
]) - (tnt - 1) * m
assert min(nt) >= 0, "Error: " + str(t)
# And the PCFG prior *not* counting t
q = lognormalize(
numpy.array([grammar.log_probability(h.value)
for h in hypotheses]) - tlp * m)
# The function to optimize
def fnc(p):
if p <= 0. or p >= 1.: return float("inf") # enforce bounds
newprior = lognormalize(q + log(p) * m + log(1. - p) * nt)
kl = 0.0
for P, kl0 in zip(Ps, KL0s):
kl += sum(numpy.exp(newprior) * (newprior - P)) / kl0
return kl
### TODO: This optimization should be analytically tractable...
### but we need to check that it is convex! Any ideas?
o = scipy.optimize.fmin(fnc,
numpy.array([0.1]),
xtol=0.0001,
ftol=0.0001,
disp=0)
print fnc(o[0]), o[0], log(o[0]), grammar.log_probability(t), qq(t)
from LOTlib import break_ctrlc
from LOTlib.TopN import TopN
from LOTlib.Inference.Samplers.MetropolisHastings import MHSampler
from Model import *
from TargetConcepts import TargetConcepts
NDATA = 20 # How many data points for each function?
NSTEPS = 100000
BEST_N = 500 # How many from each hypothesis to store
# Where we keep track of all hypotheses (across concepts)
all_hypotheses = TopN(N=BEST_N)
if __name__ == "__main__":
# Now loop over each target concept and get a set of hypotheses
for i, f in enumerate(TargetConcepts):
# Set up the hypothesis
h0 = make_hypothesis()
# Set up some data
data = make_data(NDATA, f)
# Now run some MCMC
fs = TopN(N=BEST_N, key="posterior_score")
fs.add(break_ctrlc(MHSampler(h0, data, steps=NSTEPS, trace=False)))
all_hypotheses.update(fs)
pickle.dump(all_hypotheses, open("hypotheses.pkl", 'w'))
# Hypothesis
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
from LOTlib.Hypotheses.RationalRulesLOTHypothesis import RationalRulesLOTHypothesis
def make_hypothesis(grammar=grammar, **kwargs):
return RationalRulesLOTHypothesis(grammar=grammar, rrAlpha=1.0, **kwargs)
if __name__ == "__main__":
from LOTlib.TopN import TopN
hyps = TopN(N=1000)
from LOTlib.Inference.Samplers.MetropolisHastings import MHSampler
from LOTlib import break_ctrlc
mhs = MHSampler(make_hypothesis(),
make_data(),
1000000,
likelihood_temperature=1.,
prior_temperature=1.)
for samples_yielded, h in break_ctrlc(enumerate(mhs)):
h.ll_decay = 0.
hyps.add(h)
import pickle
with open('HypothesisSpace.pkl', 'w') as f:
pickle.dump(hyps, f)
pickle.dump(hyps, f)
return hyps
###################################################################################
# Main Running
###################################################################################
argarray = map(lambda x: [x], options.data_pts * options.chains)
if is_master_process():
display_option_summary(options)
seen = set()
for fs in break_ctrlc(
MPI_map(run, numpy.random.permutation(argarray), progress_bar=False)):
for h in fs.get_all():
if h not in seen:
seen.add(h)
if h.prior > -Infinity:
print h.prior, \
h.likelihood, \
h \
#sys.stdout.flush()
#sys.stdout.flush()
import pickle
with open(options.out_path, 'w') as f:
pickle.dump(seen, f)
grammar.start = 'TWO_CONCEPT_START'
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Hypothesis
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
from Model import make_hypothesis
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Main
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
if __name__ == "__main__":
from LOTlib import break_ctrlc
from LOTlib.Inference.Samplers.MetropolisHastings import MHSampler
from LOTlib.Miscellaneous import q
# Create an initial hypothesis
# This is where we set a number of relevant variables -- whether to use RR, alpha, etc.
# Here we give args as "concept" (used in TWO_CONCEPT_START above) and "x"
h0 = make_hypothesis(grammar=grammar, args=['concept', 'x'])
data = make_data()
# Run the vanilla sampler. Without steps, it will run infinitely
# this prints out posterior (posterior_score), prior, likelihood,
for h in break_ctrlc(MHSampler(h0, data, 10000, skip=100)):
print h.posterior_score, h.prior, h.likelihood, q(h)
log(before_same_children) - log(nrk)) + old_lp_below
return [newt, f - b]
if __name__ == "__main__":
from LOTlib import break_ctrlc
#from LOTlib.Examples.Number.Shared import grammar, make_h0, generate_data
#data = generate_data(300)
## NOTE: TO NORMALLY USE THIS, YOU MUST MIX WITH REGENERATION PROPOSAL -- ELSE NOT ERGODIC
from LOTlib.Examples.Magnetism.Simple.Run import grammar, make_h0, data
from LOTlib.Inference.Samplers.MetropolisHastings import MHSampler
idp = InsertDeleteProposal(grammar)
#data = generate_data(100)
h = make_h0(proposal_function=idp)
for h in break_ctrlc(MHSampler(h, data, 100000)):
print h.posterior_score, h
"""
for _ in xrange(100):
t = grammar.generate()
print "\n\n", t
for _ in xrange(10):
print "\t", idp.propose_tree(t)
"""
def make_hypothesis(**kwargs):
return MyHypothesis(grammar=grammar, rrAlpha=1.0, **kwargs)
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Main
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
if __name__ == "__main__":
from LOTlib import break_ctrlc
from LOTlib.Inference.Samplers.MetropolisHastings import MHSampler
from LOTlib.Miscellaneous import q
# Create an initial hypothesis
# This is where we set a number of relevant variables -- whether to use RR, alpha, etc.Z
h0 = MyHypothesis(grammar, ll_decay=1.0, rrAlpha=1.0, args=['x'])
data = make_data()
# Run the vanilla sampler. Without steps, it will run infinitely
# this prints out posterior (posterior_score), prior, likelihood,
for h in break_ctrlc(
MHSampler(h0, data, 10000, skip=100, shortcut_likelihood=False)):
print h.posterior_score, h.prior, h.likelihood, q(h)
# This setup requires the *later* data to be upweighted, meaning that hypotheses that get
# later data wrong should be given lower likelhood. But also with the decay, the overall
# magnitude of the likelihood decreases.
print "# Created L, NYes, NTrials, and HOutput of size %s" % len(L)
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Run inference
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
from LOTlib import break_ctrlc
from LOTlib.Inference.GrammarInference.FullGrammarHypothesis import FullGrammarHypothesis
from LOTlib.Inference.Samplers.MetropolisHastings import MHSampler
h0 = FullGrammarHypothesis(counts, L, GroupLength, prior_offset, NYes, NTrials,
Output)
mhs = MHSampler(h0, [], 100000, skip=0)
for s, h in break_ctrlc(enumerate(mhs)):
print mhs.acceptance_ratio(), h.prior, h.likelihood,\
h.value['alpha'].value[0], h.value['beta'].value[0],\
h.value['prior_temperature'].value, h.value['likelihood_temperature'].value,\
'RULES',\
' '.join([str(x) for x in h.value['rulep']['BOOL'].value ]),\
' '.join([str(x) for x in h.value['rulep']['PREDICATE'].value ]),\
' '.join([str(x) for x in h.value['rulep']['START'].value ]),\
' '.join([str(x) for x in h.value['rulep']['SET'].value ])
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Run gradient ascent
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# from LOTlib import break_ctrlc
#
self.penalty = penalty
self.seen = Counter()
def next(self):
v = MHSampler.next(self)
self.seen[v] += 1
return v
def compute_posterior(self, h, data, **kwargs):
"""
Compute prior & likelihood for `h`, penalizing prior by how many samples have been generated so far.
"""
return self.seen[h] * self.penalty + h.compute_posterior(
data, **kwargs)
if __name__ == "__main__":
from LOTlib import break_ctrlc
from LOTlib.Examples.Number.Model import *
from LOTlib.Miscellaneous import q
data = make_data(500)
h0 = NumberExpression(grammar)
tmc = TabooMCMC(h0, data, steps=10000)
for h in break_ctrlc(tmc):
print tmc.seen[h], h.posterior_score, h.prior, h.likelihood, q(h)