language_model.py

"""
This code is modified from Hengyuan Hu's repository.
https://github.com/hengyuan-hu/bottom-up-attention-vqa
"""
import torch
import torch.nn as nn
import numpy as np

from fc import FCNet


class WordEmbedding(nn.Module):
    """Word Embedding

    The ntoken-th dim is used for padding_idx, which agrees *implicitly*
    with the definition in Dictionary.
    """
    def __init__(self, ntoken, emb_dim, dropout, op):
        super(WordEmbedding, self).__init__()
        self.op = op
        self.emb = nn.Embedding(ntoken+1, emb_dim, padding_idx=ntoken)
        if 'c' in op:
            self.emb_ = nn.Embedding(ntoken+1, emb_dim, padding_idx=ntoken)
            self.emb_.weight.requires_grad = False # fixed
        self.dropout = nn.Dropout(dropout)
        self.ntoken = ntoken
        self.emb_dim = emb_dim

    def init_embedding(self, np_file, tfidf=None, tfidf_weights=None):
        weight_init = torch.from_numpy(np.load(np_file))
        assert weight_init.shape == (self.ntoken, self.emb_dim)
        self.emb.weight.data[:self.ntoken] = weight_init
        if tfidf is not None:
            if 0 < tfidf_weights.size:
                weight_init = torch.cat([weight_init, torch.from_numpy(tfidf_weights)], 0)
            weight_init = tfidf.matmul(weight_init) # (N x N') x (N', F)
            self.emb_.weight.requires_grad = True
        if 'c' in self.op:
            self.emb_.weight.data[:self.ntoken] = weight_init.clone()

    def forward(self, x):
        emb = self.emb(x)
        if 'c' in self.op:
            emb = torch.cat((emb, self.emb_(x)), 2)
        emb = self.dropout(emb)
        return emb


class QuestionEmbedding(nn.Module):
    def __init__(self, in_dim, num_hid, nlayers, bidirect, dropout, rnn_type='GRU'):
        """Module for question embedding
        """
        super(QuestionEmbedding, self).__init__()
        assert rnn_type == 'LSTM' or rnn_type == 'GRU'
        rnn_cls = nn.LSTM if rnn_type == 'LSTM' else nn.GRU if rnn_type == 'GRU' else None

        self.rnn = rnn_cls(
            in_dim, num_hid, nlayers,
            bidirectional=bidirect,
            dropout=dropout,
            batch_first=True)

        self.in_dim = in_dim
        self.num_hid = num_hid
        self.nlayers = nlayers
        self.rnn_type = rnn_type
        self.ndirections = 1 + int(bidirect)

    def init_hidden(self, batch):
        # just to get the type of tensor
        weight = next(self.parameters()).data
        hid_shape = (self.nlayers * self.ndirections, batch, self.num_hid)
        if self.rnn_type == 'LSTM':
            return (weight.new(*hid_shape).zero_(),
                    weight.new(*hid_shape).zero_())
        else:
            return weight.new(*hid_shape).zero_()

    def forward(self, x):
        # x: [batch, sequence, in_dim]
        batch = x.size(0)
        hidden = self.init_hidden(batch)
        output, hidden = self.rnn(x, hidden)

        if self.ndirections == 1:
            return output[:, -1]

        forward_ = output[:, -1, :self.num_hid]
        backward = output[:, 0, self.num_hid:]
        return torch.cat((forward_, backward), dim=1)

    def forward_all(self, x):
        # x: [batch, sequence, in_dim]
        batch = x.size(0)
        hidden = self.init_hidden(batch)
        output, hidden = self.rnn(x, hidden)
        return output


class RnnQuestionEmbedding(nn.Module):
    def __init__(self, ntoken, w_dim, q_dim, op):
        super(RnnQuestionEmbedding, self).__init__()
        self.w_emb = WordEmbedding(ntoken, w_dim, .0, op)
        self.rnn = QuestionEmbedding(w_dim if 'c' not in op else w_dim * 2, q_dim, 1, False, .0)

    def forward(self, q):
        w_emb = self.w_emb(q)
        q_emb = self.rnn.forward_all(w_emb)
        return q_emb


class BertRnnQuestionEmbedding(nn.Module):
    def __init__(self, bert, rot_dim, q_dim, op):
        super(BertRnnQuestionEmbedding, self).__init__()
        w_dim = 768
        self.w_emb = bert
        self.w_emb_ = FCNet([w_dim, rot_dim], 'ReLU', 0.)
        self.rnn = QuestionEmbedding(rot_dim if 'c' not in op else rot_dim * 2, q_dim, 1, False, .0)

    def forward(self, q):
        w_emb, sentence_embedding = self.w_emb(q, output_all_encoded_layers=False)  # [batch, q_len, q_dim]
        q_emb = self.rnn.forward_all(self.w_emb_(w_emb))
        return q_emb