def __init__(self, model, fields, opts, final_prediction_score_multiper, alternate_model_score_multiplier, cuda=False):
self.model = model
self.fields = fields
self.copy_attn = opts.copy_attn
self.copy_attn = False
self.src_field = self.fields["src"].base_field
self.tgt_field = self.fields["tgt"].base_field
self.src_vocab = self.fields["src"].base_field.vocab
self.tgt_vocab = self.fields["tgt"].base_field.vocab
self._tgt_pad_idx = self.tgt_vocab.stoi[self.tgt_field.pad_token]
self.tt = torch.cuda if cuda else torch
self.prediction_score_multiper = final_prediction_score_multiper
self.alternate_model_score_multiplier = alternate_model_score_multiplier # S|T score multiplier
# print("final_prediction_score_multiper:", self.prediction_score_multiper)
# print("alternate_model_score_multiplier:", self.alternate_model_score_multiplier)
# self.gamma = 0.0 # Length penalty multipler
## Some additional parameters from golbal scorer just to preserve consistency
self.alpha = 0.0
self.beta = 0.0
penalty_builder = penalties.PenaltyBuilder('none', 'none')
self.has_cov_pen = penalty_builder.has_cov_pen
# Term will be subtracted from probability
self.cov_penalty = penalty_builder.coverage_penalty
self.has_len_pen = penalty_builder.has_len_pen
# Probability will be divided by this
self.length_penalty = penalty_builder.length_penalty
def __init__(self, alpha, beta, cov_penalty, length_penalty):
self.alpha = alpha
self.beta = beta
penalty_builder = penalties.PenaltyBuilder(cov_penalty, length_penalty)
# Term will be subtracted from probability
self.cov_penalty = penalty_builder.coverage_penalty()
# Probability will be divided by this
self.length_penalty = penalty_builder.length_penalty()
