def make_task_file2(aid='1834',
sizes=[50, 75, 100],
test=50,
repeats=2,
params=[{}, {}, {}],
selectsamplers=[0, 1, 2]):
'''
we drop lots of these in 1 file:
task = namedtuple("task",'samplerid size repeat sampler neg pos')
'''
pos, neg = util.getgraphs(aid)
tasks = []
for size in sizes:
repeatsXposnegsamples = util.sample_pos_neg(pos, neg, size + test,
size + test, repeats)
repeatsXposnegsamples = [[
pos2[:size], neg2[:size], pos2[size:], neg2[size:]
] for pos2, neg2 in repeatsXposnegsamples]
for i, sampler in enumerate(
util.get_all_samplers(params=params, select=selectsamplers)):
for j, (pos_sample, neg_sample, pos_test,
neg_test) in enumerate(repeatsXposnegsamples):
tasks.append(
util.task2(i, size, j, sampler, neg_sample, pos_sample,
neg_test, pos_test))
util.dumpfile(tasks, "natlearn/t2_%s_%d" % (aid, max(sizes)))
return "t2_%s_%d" % (aid, max(sizes))
def make_task_file(aid='1834',
sizes=[50, 75, 100],
repeats=2,
params=[{}, {}, {}],
selectsamplers=[0, 1, 2],
taskfile_poststring=''):
'''
we drop lots of these in 1 file:
task = namedtuple("task",'samplerid size repeat sampler neg pos')
'''
pos, neg = util.getgraphs(aid)
tasks = []
models = []
for size in sizes:
repeatsXposnegsamples, estis = util.sample_pos_neg_ESTI(
pos, neg, size, size, repeats)
models.append(estis)
for i, sampler in enumerate(
util.get_all_samplers(params=params, select=selectsamplers)):
for j, (pos_sample,
neg_sample) in enumerate(repeatsXposnegsamples):
tasks.append(
util.task(i, size, j, sampler, neg_sample, pos_sample))
fname = "%s_%d%s" % (aid, max(sizes), taskfile_poststring)
util.dumpfile(tasks, fname)
util.dumpfile(models, fname + "_models")
return fname
def run(fname, idd):
task = util.loadfile(fname)[idd]
#draw.graphlearn(decomposers[:5],size=10)
esti = util.graphs_to_linmodel(task.pos, task.neg)
X, y = util.graphs_to_Xy(task.postest, task.negtest)
ypred = esti.predict(X)
acc = sklearn.metrics.accuracy_score(y, ypred)
util.dumpfile((task.size, acc), "ASD/d_%s_%d" % (fname, idd))
def make_task(aid, sizes=[50,100,150], test=100,repeats=3, params={},postname=''):
pos,neg = util.getgraphs(aid)
tasks=[]
for size in sizes:
getsize=size+test
for i,(p,n) in enumerate(util.sample_pos_neg(pos,neg,getsize,getsize,repeats)):
sampler = util.get_casc_abstr(kwargs=params)
tasks.append(util.task2(1,size,i,sampler,n[:size],p[:size],n[size:],p[size:]))
util.dumpfile(tasks,"stack_task"+postname)
def run(filename, taskid):
tasks = util.loadfile(filename)
try:
result = util.sample(tasks[taskid])
except Exception as exc:
print "molelearnedlayer except"
#print tasks[taskid]
#import traceback
#print traceback.format_exc(20)
return None
util.dumpfile(result, getresfilename(filename, taskid))
def run(filename, taskid):
tasks = util.loadfile("natlearn/" + filename)
task = gettask(tasks, taskid)
try:
result = util.sample(task)
except Exception as exc:
print "naturallearning is showing the task object:"
print task
import traceback
print traceback.format_exc(20)
return None
util.dumpfile(result, "natlearn/res_%s_%d" % (filename, taskid))
def evalandshow(fname, tasknum, sizes, show=False):
#oracle = util.aid_to_linmodel(aid)
res = readresults(fname, tasknum)
processed = eval(res, util.loadfile(fname + "_models"), sizes)
util.dumpfile((processed, fname, show), "ASASD")
draw2(processed, fname + "score.png", show=show)
draw(processed,
fname + "time.png",
get_mean=lambda x: x.time_mean,
get_var=lambda x: x.time_var,
show=show)
def run(fname,idd):
def getacc(esti, a,b):
X,y = util.graphs_to_Xy(a,b)
ypred = esti.predict(X)
acc = sklearn.metrics.accuracy_score(y,ypred)
return acc
task = util.loadfile(fname)[idd]
# make an estimator with the full stack
full_stacked = [task.sampler.decomposer.make_new_decomposer(data).pre_vectorizer_graph()
for data in task.sampler.graph_transformer.fit_transform(task.pos,task.neg,
remove_intermediary_layers=False)]
esti= util.graphs_to_linmodel( full_stacked[:task.size], full_stacked[task.size:] )
# fully stacked test instances
testgraphs= [task.sampler.decomposer.make_new_decomposer(data).pre_vectorizer_graph()
for data in task.sampler.graph_transformer.transform(task.postest+task.negtest,remove_intermediary_layers=False)]
acc= getacc(esti, testgraphs[:len(task.postest)], testgraphs[len(task.postest):] )
util.dumpfile((task.size,acc), "stacked/%s_%d" % (fname,idd))
task = util.loadfile(fname)[idd]
# make an estimator with the full stack
full_stacked = [task.sampler.decomposer.make_new_decomposer(data).pre_vectorizer_graph()
for data in task.sampler.graph_transformer.fit_transform(task.pos,task.neg,
remove_intermediary_layers=True)]
esti= util.graphs_to_linmodel( full_stacked[:task.size], full_stacked[task.size:] )
# fully stacked test instances
testgraphs= [task.sampler.decomposer.make_new_decomposer(data).pre_vectorizer_graph()
for data in task.sampler.graph_transformer.transform(task.postest+task.negtest,remove_intermediary_layers=True)]
acc= getacc(esti, testgraphs[:len(task.postest)], testgraphs[len(task.postest):] )
util.dumpfile((task.size,acc), "stacked/2_%s_%d" % (fname,idd))
