print("betaMomentum=", betaMomentum)
print("numIterFixed=", numIterFixed)
print("maxIter=", maxIter)
print("minibatchSize=", minibatchSize)
multitask = False
print("multitask=", multitask)
if not (multitask):
numTasks = 1
for ind in trainingDataNegSampling.index:
a = deepcopy(trainingDataNegSampling.loc[ind, 'itemSet'])
a = a + [trainingDataNegSampling.loc[ind, 'target']]
trainingDataNegSampling.at[ind, 'itemSet'] = a
mod = logisticMultiTaskDPP(setName, taskName, rewardName, numItems, numTasks,
numTraits, lbda, alpha, eps, betaMomentum,
numIterFixed, minibatchSize, maxIter, gradient_cap,
random_state)
mod.fit(trainingDataNegSampling, testData_sample)
#np.savetxt(path+"V.txt",mod.V)
#np.savetxt(path+"D.txt",mod.D)
#
#R = np.zeros((numTasks,numTraits))
#for task in range(numTasks):
# R[task,:] = mod.R[task]
#np.savetxt(path+"R.txt",R)
print("End learning")
print("numTraits:", numTraits)
print("minibatchSize:", minibatchSize)
np.random.seed(123*run)
data['cv'] = np.random.random(len(data))
trainingData = data.loc[data['cv']<threshold,]
testData = data.loc[data['cv']>=threshold,]
testData = testData.loc[(testData['setSize']>1) & (testData['conversion']==1),]
testData['target2'] = -1
for ind in testData.index:
testData.loc[ind,'target2'] = testData.loc[ind,'itemSet'][-1]
testData.at[ind,'itemSet'] = testData.loc[ind,'itemSet'][:-1]
mod = logisticMultiTaskDPP(setName=setName, taskName=taskName,
rewardName=rewardName, numItems=numItems,
numTasks=numTasks, numTraits=numTraits,
lbda=lbda, alpha=alpha, eps=eps,
betaMomentum=betaMomentum, numIterFixed=numIterFixed,
minibatchSize=minibatchSize, maxIter=maxIter,
gradient_cap=gradient_cap, random_state=random_state,
verbose=False)
nCat = len(category)
numTraits_cat = int(numTraits/nCat)
nItem_cat = int(numItems/nCat)
V0 = np.random.normal(scale=0.001,size=(numItems,numTraits))
R0 = {}
for i in range(len(category)):
V0[nItem_cat*i:(nItem_cat*(i+1)),numTraits_cat*i:(numTraits_cat*(i+1))] *= 10
for task in range(i*nItem_cat,(i+1)*nItem_cat):
R0[task] = np.random.normal(loc=0.0,scale=0.01,size=numTraits)
R0[task][numTraits_cat*i:(numTraits_cat*(i+1))] += 1.0