def __init__(self, bert_large=False):
super().__init__(bert_large)
MUS = [-0.9, -0.7, -0.5, -0.3, -0.1, 0.1, 0.3, 0.5, 0.7, 0.9, 1.0]
SIGMAS = [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.001]
self.bert_ranker = VanillaBertRanker()
self.simmat = modeling_util.SimmatModule()
self.kernels = modeling_util.KNRMRbfKernelBank(MUS, SIGMAS)
self.combine = torch.nn.Linear(self.kernels.count() * self.CHANNELS + self.BERT_SIZE, 1)
def __init__(self, bert_large=False):
super().__init__(bert_large)
NBINS = 11
HIDDEN = 5
self.bert_ranker = VanillaBertRanker()
self.simmat = modeling_util.SimmatModule()
self.histogram = modeling_util.DRMMLogCountHistogram(NBINS)
self.hidden_1 = torch.nn.Linear(NBINS * self.CHANNELS + self.BERT_SIZE, HIDDEN)
self.hidden_2 = torch.nn.Linear(HIDDEN, 1)
def __init__(self, bert_large=False):
super().__init__(bert_large)
QLEN = 20
KMAX = 2
NFILTERS = 32
MINGRAM = 1
MAXGRAM = 3
self.simmat = modeling_util.SimmatModule()
self.ngrams = torch.nn.ModuleList()
self.rbf_bank = None
for ng in range(MINGRAM, MAXGRAM+1):
ng = modeling_util.PACRRConvMax2dModule(ng, NFILTERS, k=KMAX, channels=self.CHANNELS)
self.ngrams.append(ng)
qvalue_size = len(self.ngrams) * KMAX
self.linear1 = torch.nn.Linear(self.BERT_SIZE + QLEN * qvalue_size, 32)
self.linear2 = torch.nn.Linear(32, 32)
self.linear3 = torch.nn.Linear(32, 1)