def getExamples(con, experimentName, callback, callbackArgs, metaDataFileName=None, options=None, experimentMeta=None):
con = connect(con)
template = parseExperiment(experimentName).copy()
template = parseTemplateOptions(options, template)
#con = connect(con, template.get("functions", None))
#options = updateTemplateOptions(template, options)
print "Template:", experimentName
print json.dumps(template, indent=4)
compiled, lambdaArgs = compileTemplate(template)
print "Compiled experiment"
examples = [dict(x) for x in compiled["example"](con=con, **lambdaArgs)]
numHidden = hidden.setHiddenValues(examples, compiled)
numExamples = len(examples)
print "Examples " + str(numExamples) + ", hidden " + str(numHidden)
count = 0
clsIds = compiled.get("classes", None)
hiddenRule = compiled.get("include", "train")
featureIds = {}
meta = []
featureGroups = compiled.get("features", [])
sampleRandom = MTwister()
sampleRandom.set_seed(2)
for example in examples:
count += 1
if not hidden.getInclude(example, compiled.get("hidden", None), hiddenRule):
continue
hidden.setSet(example, compiled.get("hidden", None))
#print experiment["class"](con, example)
#if count % 10 == 0:
print "Processing example", example,
cls = getIdOrValue(compiled["label"](con=con, example=example, **lambdaArgs), clsIds)
print cls, str(count) + "/" + str(numExamples)
strCls = str(cls)
if "sample" in compiled and strCls in compiled["sample"] and sampleRandom.random() > compiled["sample"][strCls]:
print "NOTE: Downsampled example"
continue
if "filter" in compiled and compiled["filter"] != None and len([x for x in compiled["filter"](con=con, example=example, **lambdaArgs)]) == 0:
print "NOTE: Filtered example"
continue
features = {}
for featureGroup in featureGroups:
for row in featureGroup(con=con, example=example, **lambdaArgs):
for key, value in itertools.izip(*[iter(row)] * 2): # iterate over each consecutive key,value columns pair
if not isinstance(key, basestring):
raise Exception("Non-string feature key '" + str(key) + "' in feature group " + str(featureGroups.index(featureGroup)))
if not isinstance(value, Number):
raise Exception("Non-number feature value '" + str(value) + "' in feature group " + str(featureGroups.index(featureGroup)))
features[getId(key, featureIds)] = value
if len(features) == 0:
print "WARNING: example has no features"
if callback != None:
callback(example=example, cls=cls, features=features, **callbackArgs)
if "meta" in compiled:
meta.append(compiled["meta"](label=cls, features=features, example=example, **lambdaArgs))
saveMetaData(metaDataFileName, con, template, experimentName, options, clsIds, featureIds, meta, experimentMeta)
return featureIds
class HiddenSet():
def __init__(self, seed=1):
self.__random = MTwister()
self.__random.set_seed(seed)
self.__thresholds = []
def getThreshold(self, index):
while len(self.__thresholds) <= index:
self.__thresholds.append(self.__random.random())
return self.__thresholds[index]
def getDonorThreshold(self, donorId):
donorIndex = int(donorId[2:])
return self.getThreshold(donorIndex)
"""
For calculating a hidden set of donors.
"""
import sys, os
import result
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from lib.pymersennetwister.mtwister import MTwister
__hiddenRandom = MTwister()
__hiddenRandom.set_seed(1)
__hidden = []
def getHiddenValue(index):
while len(__hidden) <= index:
__hidden.append(__hiddenRandom.random())
return __hidden[index]
def getDonorHiddenValue(icgc_donor_id):
return getHiddenValue(int(icgc_donor_id[2:]))
def setHiddenValues(examples, template, donorIdKey="icgc_donor_id"):
numHidden = 0
if "hidden" in template:
for example in examples:
example["hidden"] = getDonorHiddenValue(example[donorIdKey])
if example["hidden"] < template["hidden"]:
numHidden += 1
return numHidden
def getInclude(example, templateHidden, hiddenRule, verbose=True):
if hiddenRule not in ("train", "hidden", "both"):
def __init__(self, seed=1):
self.__random = MTwister()
self.__random.set_seed(seed)
self.__thresholds = []
