EncDecModel.py

import torch.nn as nn
from Utils import *
from modules.Generations import *


class EncDecModel(nn.Module):
    def __init__(self, vocab2id, max_dec_len=120, beam_width=1, eps=1e-10):
        super(EncDecModel, self).__init__()
        self.eps = eps
        self.beam_width = beam_width
        self.max_dec_len = max_dec_len
        self.vocab2id = vocab2id

    def encode(self, data):
        raise NotImplementedError

    def init_decoder_states(self, data, encode_output):
        return None

    def init_feedback_states(self, encode_outputs, init_decoder_states):
        return None

    def decode(self, data, previous_word, encode_outputs, previous_deocde_outputs, feedback_outputs):
        raise NotImplementedError

    def generate(self, data, encode_outputs, decode_outputs, softmax=False):
        raise NotImplementedError

    def to_word(self, data, gen_outputs, k=5, sampling=False):
        raise NotImplementedError

    def generation_to_decoder_input(self, data, indices):
        return indices

    def decoder_to_encoder(self, data, encoder_outputs, decoder_outputs):
        return NotImplementedError

    def loss(self, data, encode_output, decode_outputs, gen_outputs, reduction='mean'):
        raise NotImplementedError

    def to_sentence(self, data, batch_indice):
        raise NotImplementedError

    def sample(self, data):
        raise NotImplementedError

    def greedy(self, data):
        return greedy(self, data, self.vocab2id, self.max_dec_len)

    def beam(self, data):
        return beam(self, data, self.vocab2id, self.max_dec_len, self.beam_width)