def fit_sample_noabstr(sequences, argz, random_state=random.random()):
'''
graphs -> more graphs
graphs are pretty mich (NAME,SEQUENCE),()...
'''
# fit a sampler
sequences = list(sequences)
estimator = estimatorwrapper(
nu=.5, cv=2, n_jobs=1) # with .5 it also works for the fewer ones..
sampler = rna.AbstractSampler(
radius_list=argz['radius_list'], #[0, 1, 2], # war 0,1
thickness_list=argz['thickness_list'], #[1], # war 2
min_cip_count=argz['mincip_count'],
min_interface_count=argz['min_interfacecount'],
preprocessor=rnana.PreProcessor(base_thickness_list=[1],
ignore_inserts=True),
postprocessor=rna.PostProcessor(),
estimator=estimator
# feasibility_checker=feasibility
)
sampler.fit(sequences, grammar_n_jobs=1, grammar_batch_size=1)
# logger.info('graph grammar stats:')
dataset_size, interface_counts, core_counts, cip_counts = sampler.grammar(
).size()
# logger.info('#instances:%d #interfaces: %d #cores: %d #core-interface-pairs: %d' % (dataset_size, interface_counts, core_counts, cip_counts))
sequences = [b for a, b in sequences]
sequences = sampler.sample(sequences,
n_samples=5,
batch_size=1,
n_steps=55,
n_jobs=1,
quick_skip_orig_cip=True,
probabilistic_core_choice=False,
burnin=6,
improving_threshold=0.5,
improving_linear_start=0.15,
max_size_diff=6,
accept_min_similarity=0.55,
select_cip_max_tries=30,
keep_duplicates=False,
include_seed=False,
backtrack=2,
monitor=False)
result = []
for li in sequences:
result += li
return [r[1] for r in result]
def oneclasstest_fraction(fraction=0.1, repeats=2):
# choosing some graphs,
# having array to save results
for i in range(repeats):
badscores = []
goodscores = []
graphs = get_sequences_with_names(size=923)
graphs, not_used = random_bipartition_iter(
graphs, fraction, random_state=random.random() * i * 1000)
estimator = Wrapper(nu=.27, cv=3, n_jobs=-1)
sampler = rna.AbstractSampler(radius_list=[0, 1],
thickness_list=[2],
min_cip_count=1,
min_interface_count=2,
preprocessor=rna.PreProcessor(
base_thickness_list=[1],
ignore_inserts=True),
postprocessor=rna.PostProcessor(),
estimator=estimator)
sampler.preprocessor.set_param(sampler.vectorizer)
graphmanagers = sampler.preprocessor.fit_transform(graphs)
sampler.estimatorobject.fit(graphmanagers,
vectorizer=sampler.vectorizer,
random_state=sampler.random_state)
#test
for graphman in graphmanagers:
struct = evaltools.dotbracket_to_shape(graphman.structure,
shapesversion=SHAPEVERSION)
score = sampler.estimatorobject.score(graphman)
if struct == "[[][][]]":
goodscores.append(score)
else:
badscores.append(score)
print "afraction=%f , instances=%f, good=%d , bad=%d" % (
fraction, fraction * 923, len(goodscores), len(badscores))
a = numpy.array(badscores)
print 'bad:mean/std ', numpy.mean(a, axis=0), ' ', numpy.std(a, axis=0)
a = numpy.array(goodscores)
print 'cgood:mean/std ', numpy.mean(a, axis=0), ' ', numpy.std(a,
axis=0)
a = numpy.array(goodscores + badscores)
print 'dbad+good:mean/std ', numpy.mean(a,
axis=0), ' ', numpy.std(a,
axis=0)
print ''
def fit_sample(graphs, random_state=random.random()):
'''
graphs -> more graphs
'''
graphs = list(graphs)
estimator = estimatorwrapper(nu=.5, cv=2, n_jobs=-1)
sampler = rna.AbstractSampler(radius_list=[0, 1],
thickness_list=[2],
min_cip_count=1,
min_interface_count=2,
preprocessor=rna.PreProcessor(
base_thickness_list=[1],
ignore_inserts=True),
postprocessor=rna.PostProcessor(),
estimator=estimator
#feasibility_checker=feasibility
)
sampler.fit(graphs, grammar_n_jobs=4, grammar_batch_size=1)
logger.info('graph grammar stats:')
dataset_size, interface_counts, core_counts, cip_counts = sampler.grammar(
).size()
logger.info(
'#instances:%d #interfaces: %d #cores: %d #core-interface-pairs: %d'
% (dataset_size, interface_counts, core_counts, cip_counts))
graphs = [b for a, b in graphs]
graphs = sampler.sample(graphs,
n_samples=3,
batch_size=1,
n_steps=50,
n_jobs=4,
quick_skip_orig_cip=True,
probabilistic_core_choice=True,
burnin=10,
improving_threshold=0.9,
improving_linear_start=0.3,
max_size_diff=20,
accept_min_similarity=0.65,
select_cip_max_tries=30,
keep_duplicates=False,
include_seed=False,
backtrack=10,
monitor=False)
result = []
for graphlist in graphs:
result += graphlist
# note that this is a list [('',sequ),..]
return result
def fit_sample(graphs, random_state=random.random()):
'''
graphs -> more graphs
arguments are generated above Oo
'''
global arguments
graphs = list(graphs)
estimator = estimatorwrapper(nu=.5, cv=2, n_jobs=NJOBS)
sampler = rna.AbstractSampler(
radius_list=[0, 1],
thickness_list=[2],
min_cip_count=arguments['mincipcount'],
min_interface_count=arguments['mininterfacecount'],
preprocessor=rna.PreProcessor(base_thickness_list=[1],
ignore_inserts=True),
postprocessor=rna.PostProcessor(),
estimator=estimator
#feasibility_checker=feasibility
)
sampler.fit(graphs, grammar_n_jobs=NJOBS, grammar_batch_size=1)
graphs = [b for a, b in graphs]
graphs = sampler.sample(graphs,
n_samples=arguments['n_samples'],
batch_size=1,
n_steps=arguments['n_steps'],
n_jobs=NJOBS,
quick_skip_orig_cip=arguments['quick_skip'],
probabilistic_core_choice=arguments['core_choice'],
burnin=arguments['burnin'],
improving_threshold=arguments['imp_thresh'],
improving_linear_start=arguments['imp_lin_start'],
max_size_diff=arguments['maxsizediff'],
accept_min_similarity=arguments['acc_min_sim'],
select_cip_max_tries=30,
keep_duplicates=False,
include_seed=False,
backtrack=2,
monitor=False)
result = []
for graphlist in graphs:
result += graphlist
# note that this is a list [('',sequ),..]
return result
def fit_sample_infernal(seques, dummy):
"""
ok wir machen
write fasta
muscle,
alifold,
biopython,
create_cm und cmemit oO
"""
#print seques
sequences = [b for a, b in seques]
rna.write_fasta(sequences, "tmp.fa")
shell_exec('muscle -in tmp.fa -out museld.fa')
a, b, out = shell_exec('cat museld.fa | RNAalifold -f F --noPS')
ss = getstr(out)
to_stockholm('museld.fa', ss, 'sto.sto')
shell_exec("cmbuild -F mod3l sto.sto")
shell_exec("cmemit -N %d --exp 3.92 mod3l > out.fa" %
(len(sequences) * 2))
return fasta_to_list('out.fa')
def eval(repeats,size):
result=[]
for i in range(repeats):
graphs=get_sequences_with_names(size=size, rand=(i+3)*10)
zz=fit_sample(graphs)
z=[b for a ,b in zz]
cmpath='../%s.cm' % RFAM
result+=rna.infernal_checker(z,cmfile=cmpath, cmsearchbinarypath='../toolsdata/cmsearch')
a = numpy.array(result)
mean = numpy.mean(a, axis=0)
std = numpy.std(a, axis=0)
print 'size:%d mean:%f std:%f' % (size,mean,std)
return mean,std
def eval(repeats, size):
result = []
for i in range(repeats):
graphs = get_sequences_with_names(size=size, rand=(i + 3) * 10)
zz = fit_sample(graphs)
z = [b for a, b in zz]
cmpath = '../%s.cm' % RFAM
result += rna.infernal_checker(
z, cmfile=cmpath, cmsearchbinarypath='../toolsdata/cmsearch')
a = numpy.array(result)
mean = numpy.mean(a, axis=0)
std = numpy.std(a, axis=0)
print 'size:%d mean:%f std:%f' % (size, mean, std)
return mean, std
def evaluate(repeats, size, fitsample, RFAM,inputdict,debug):
means = []
stds = []
for i in range(repeats):
if debug: print 'start rep'
sequences,void = utils.get_seq_tups(RFAM+'.fa',size,1)
zz = fitsample(sequences,inputdict)
# print zz[:3]
zz=[b for a ,b in zz]
result = rna.infernal_checker(zz, cmfile='../toolsdata/%s.cm' % RFAM,
cmsearchbinarypath='../toolsdata/cmsearch')
a = np.array(result)
means.append(np.mean(a, axis=0))
stds.append(np.std(a, axis=0))
means.sort()
stds.sort()
#print (size, means, stds)
return means[repeats / 2] * 100, stds[repeats / 2] * 100
def evaluate(repeats, size, fitsample):
print 'eval:',
means = []
stds = []
for i in range(repeats):
sequences = get_sequences_with_names(size=size, rand=10)
zz = fitsample(sequences)
# print zz[:3]
# z=[b for a ,b in zz]
result = rna.infernal_checker(
zz,
cmfile='../toolsdata/%s.cm' % RFAM,
cmsearchbinarypath='../toolsdata/cmsearch')
a = np.array(result)
means.append(np.mean(a, axis=0))
stds.append(np.std(a, axis=0))
means.sort()
stds.sort()
print(size, means, stds)
return [means[repeats / 2] * 100, stds[repeats / 2] * 100]
sequences = itertools.islice(fasta_to_sequence("../example/RF00005.fa"),
size)
return [b for (a, b) in sequences]
def get_sequences_with_names(size=9999):
sequences = itertools.islice(fasta_to_sequence("../example/RF00005.fa"),
size)
return sequences
'''
learning a grammar
'''
import graphlearn.abstract_graphs.learned_RNA as learned
import graphlearn.abstract_graphs.RNA as rna
from graphlearn import feasibility
feas = feasibility.FeasibilityChecker(
checklist=[feasibility.default_check, rna.is_rna])
graphs = get_sequences_with_names(150)
pp = learned.RnaPreProcessor(base_thickness_list=[2],
kmeans_clusters=3,
structure_mod=False)
sampler = rna.AbstractSampler(radius_list=[0, 1],
thickness_list=[1],
min_cip_count=2,
min_interface_count=2,
feasibility_checker=feas,
preprocessor=pp)
sampler.fit(graphs, grammar_n_jobs=1, grammar_batch_size=1)