def testBestReRank(k,bestModel=None):
if bestModel is None:
model=CascadingModel(tasktype=modeltype,digtype=datatype)
else:
model=bestModel
mymetric=model.mymetric
mymetric.verbose=0
taskids=data.taskids[:data.testPoint]
Y_label=data.testLabel[:data.testPoint]
print("tuning re-rank weight\n")
best_param={3:0,5:0,10:0}
rw=[w/10 for w in range(0,11)]
rw.reverse()
model.topK=k
maxAcc=[0,0]
model.loadConf()
model.loadModel()
Y_predict2=model.predict(data.testX,taskids)
for w in rw:
acc=mymetric.topKPDIGUsers(Y_predict2,Y_label,taskids,k,w)
acc=np.mean(acc)
if acc>maxAcc[0]:
maxAcc=[acc,w]
#print(data.tasktype,"top %d"%k,acc,"weight=%f"%(w/10))
best_param[k]=maxAcc[1]
print("\n",data.tasktype,"top %d"%k,maxAcc[0],"weight=%f"%maxAcc[1])
print()
def testCascadingModel():
model=CascadingModel(tasktype=modeltype)
taskids=data.taskids[:data.testPoint]
Y_label=data.testLabel[:data.testPoint]
Y_sublabel=data.submitLabelClassification[:data.testPoint]
Y_reglabel=data.registerLabelClassification[:data.testPoint]
mymetric=model.mymetric
mymetric.callall=False
print()
print("\n meta-learning model top k acc")
for k in (3,5,10):
model.topK=k
model.loadConf()
model.loadModel()
Y_predict2=model.predict(data.testX,taskids)
#metrics
acc=mymetric.topKPossibleUsers(Y_predict2,Y_label,k)
acc=np.mean(acc)
print(data.tasktype,"top %d"%k,acc)
acc=mymetric.topKRUsers(Y_predict2,Y_label,Y_reglabel,k,)
acc=np.mean(acc)
print(data.tasktype,"top %d"%k,acc)
acc=mymetric.topKSUsers(Y_predict2,Y_label,Y_sublabel,k,)
acc=np.mean(acc)
print(data.tasktype,"top %d"%k,acc)
print()