release

README.md

@@ -0,0 +1,49 @@
+Code for the ACL 2019 paper:
+
+## Improving Question Answering over Incomplete KBs with Knowledge-Aware Reader
+
+Paper link: [https://arxiv.org/abs/1905.07098](https://arxiv.org/abs/1905.07098)
+
+Model Overview:
+<p align="center"><img width="90%" src="assets/model.png" /></p>
+
+### Requirements
+* ``PyTorch 1.0.1``
+* ``tensorboardX``
+* ``tqdm``
+* ``gluonnlp``
+
+### Prepare data
+```
+mkdir datasets && cd datasets && wget http://nlp.cs.ucsb.edu/data/webqsp.tar.gz && tar -xzvf webqsp.tar.gz
+```
+
+### Full KB setting
+**Training**
+```
+CUDA_VISIBLE_DEVICES=0 python train.py --model_id KAReader_full_kb --num_layer 1 --max_num_neighbors 50 --label_smooth 0.1 --data_folder datasets/webqsp/full/ 
+```
+**Testing**
+```
+CUDA_VISIBLE_DEVICES=0 python train.py --model_id KAReader_full_kb --num_layer 1 --max_num_neighbors 50 --label_smooth 0.1 --data_folder datasets/webqsp/full/ --mode test
+```
+
+### Incomplete KB setting (50%)
+**Training**
+```
+CUDA_VISIBLE_DEVICES=0 python train.py --model_id KAReader_kb_05 --num_layer 1 --max_num_neighbors 100 --use_doc --label_smooth 0.1 --data_folder datasets/webqsp/kb_05/
+```
+**Testing**
+```
+CUDA_VISIBLE_DEVICES=0 python train.py --model_id KAReader_kb_05 --num_layer 1 --max_num_neighbors 100 --use_doc --label_smooth 0.1 --data_folder datasets/webqsp/kb_05/ --mode test --eps 0.12
+```
+
+### Bibtex
+```
+@article{xiong2019improving,
+  title={Improving Question Answering over Incomplete KBs with Knowledge-Aware Reader},
+  author={Xiong, Wenhan and Yu, Mo and Chang, Shiyu and Guo, Xiaoxiao and Wang, William Yang},
+  journal={arXiv preprint arXiv:1905.07098},
+  year={2019}
+}
+```

attention.py

@@ -0,0 +1,169 @@
+import torch
+import torch.nn.functional as F
+import torch.nn as nn
+from torch.autograd import Variable
+import copy
+import math
+
+def attention(query, key, value, mask=None, dropout=None):
+    "Compute 'Scaled Dot Product Attention'"
+    d_k = query.size(-1)
+    scores = torch.matmul(query, key.transpose(-2, -1)) \
+             / math.sqrt(d_k)
+    if mask is not None:
+        scores = scores.masked_fill(mask == 0, -1e9)
+    p_attn = F.softmax(scores, dim = -1)
+    if dropout is not None:
+        p_attn = dropout(p_attn)
+    return torch.matmul(p_attn, value), p_attn
+
+
+def clones(module, N):
+    "Produce N identical layers."
+    return nn.ModuleList([copy.deepcopy(module) for _ in range(N)])
+
+
+class MultiHeadedAttention(nn.Module):
+    def __init__(self, h, d_model, dropout=0.1):
+        "Take in model size and number of heads."
+        super(MultiHeadedAttention, self).__init__()
+        assert d_model % h == 0
+        # We assume d_v always equals d_k
+        self.d_k = d_model // h
+        self.h = h
+        self.linears = clones(nn.Linear(d_model, d_model), 4)
+        self.attn = None
+        self.dropout = nn.Dropout(p=dropout)
+
+    def forward(self, query, key, value, mask=None):
+        "Implements Figure 2"
+        if mask is not None:
+            # Same mask applied to all h heads.
+            mask = mask.unsqueeze(1)
+        nbatches = query.size(0)
+
+        # 1) Do all the linear projections in batch from d_model => h x d_k 
+        query, key, value = \
+            [l(x).view(nbatches, -1, self.h, self.d_k).transpose(1, 2)
+             for l, x in zip(self.linears, (query, key, value))]
+
+        # 2) Apply attention on all the projected vectors in batch. 
+        x, self.attn = attention(query, key, value, mask=mask, 
+                                 dropout=self.dropout)
+
+        # 3) "Concat" using a view and apply a final linear. 
+        x = x.transpose(1, 2).contiguous() \
+             .view(nbatches, -1, self.h * self.d_k)
+        return self.linears[-1](x)
+
+class PositionwiseFeedForward(nn.Module):
+    "Implements FFN equation."
+    def __init__(self, d_model, d_ff, dropout=0.1):
+        super(PositionwiseFeedForward, self).__init__()
+        self.w_1 = nn.Linear(d_model, d_ff)
+        self.w_2 = nn.Linear(d_ff, d_model)
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, x):
+        return self.w_2(self.dropout(F.relu(self.w_1(x))))
+
+class PositionalEncoding(nn.Module):
+    "Implement the PE function."
+    def __init__(self, d_model, dropout, max_len=5000):
+        super(PositionalEncoding, self).__init__()
+        self.dropout = nn.Dropout(p=dropout)
+
+        # Compute the positional encodings once in log space.
+        pe = torch.zeros(max_len, d_model)
+        position = torch.arange(0, max_len).unsqueeze(1).float()
+        div_term = torch.exp(torch.arange(0, d_model, 2).float() *
+                             -(math.log(10000.0) / d_model))
+        pe[:, 0::2] = torch.sin(position * div_term)
+        pe[:, 1::2] = torch.cos(position * div_term)
+        pe = pe.unsqueeze(0)
+        self.register_buffer('pe', pe)
+
+    def forward(self, x):
+        x = x + Variable(self.pe[:, :x.size(1)], 
+                         requires_grad=False)
+        return self.dropout(x)
+
+class LayerNorm(nn.Module):
+    "Construct a layernorm module (See citation for details)."
+    def __init__(self, features, eps=1e-6):
+        super(LayerNorm, self).__init__()
+        self.a_2 = nn.Parameter(torch.ones(features))
+        self.b_2 = nn.Parameter(torch.zeros(features))
+        self.eps = eps
+
+    def forward(self, x):
+        mean = x.mean(-1, keepdim=True)
+        std = x.std(-1, keepdim=True)
+        return self.a_2 * (x - mean) / (std + self.eps) + self.b_2
+
+class SublayerConnection(nn.Module):
+    """
+    A residual connection followed by a layer norm.
+    Note for code simplicity the norm is first as opposed to last.
+    """
+    def __init__(self, size, dropout):
+        super(SublayerConnection, self).__init__()
+        self.norm = LayerNorm(size)
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, x, sublayer):
+        "Apply residual connection to any sublayer with the same size."
+        return x + self.dropout(sublayer(self.norm(x)))
+
+class EncoderLayer(nn.Module):
+    "Encoder is made up of self-attn and feed forward (defined below)"
+    def __init__(self, size, self_attn, feed_forward, dropout):
+        super(EncoderLayer, self).__init__()
+        self.self_attn = self_attn
+        self.feed_forward = feed_forward
+        self.sublayer = clones(SublayerConnection(size, dropout), 2)
+        self.size = size
+
+    def forward(self, x, mask):
+        "Follow Figure 1 (left) for connections."
+        x = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, mask))
+        return self.sublayer[1](x, self.feed_forward)
+
+class Encoder(nn.Module):
+    "Core encoder is a stack of N layers"
+    def __init__(self, layer, N):
+        super(Encoder, self).__init__()
+        self.layers = clones(layer, N)
+        self.norm = LayerNorm(layer.size)
+
+    def forward(self, x, mask):
+        "Pass the input (and mask) through each layer in turn."
+        for layer in self.layers:
+            x = layer(x, mask)
+        return self.norm(x)
+
+class SimpleEncoder(nn.Module):
+    """
+    takes (batch_size, seq_len, embed_dim) as inputs
+    calculate MASK, POSITION_ENCODING 
+    """
+    def __init__(self, embed_dim, head=4, layer=1, dropout=0.1):
+        super(SimpleEncoder, self).__init__()
+        d_ff = 2 * embed_dim
+
+        self.position = PositionalEncoding(embed_dim, dropout)
+        attn = MultiHeadedAttention(head, embed_dim)
+        ff = PositionwiseFeedForward(embed_dim, d_ff)
+        self.encoder = Encoder(EncoderLayer(embed_dim, attn, ff, dropout), layer)
+
+    def forward(self, x, mask):
+        mask = mask.unsqueeze(-2)
+        x = self.position(x)
+        x = self.encoder(x, mask)
+        return x
+
+if __name__ == '__main__':
+    encoder = SimpleEncoder(350, 2, 1)
+    inputs = torch.zeros(1000,50,350)
+    lens  = [10] * 1000
+    encoder(inputs, lens)

build_emb.py

@@ -0,0 +1,73 @@
+import gluonnlp as nlp
+import numpy as np
+from tqdm import tqdm
+
+dataset = 'datasets/webqsp/kb_05'
+rel_path = dataset + '/relations.txt'
+
+word_counter = []
+
+# load original vocab
+with open(dataset + '/vocab.txt') as f:
+    for line in f.readlines():
+        word_counter.append(line.strip())
+
+rel_words = []
+max_num_words = 0
+all_relations = []
+
+# how to split the relation
+if 'webqsp' in dataset:
+    with open(rel_path) as f:
+        first_line = True
+        for line in tqdm(f.readlines()):
+            if first_line:
+                first_line = False
+                continue
+            line = line.strip()
+            all_relations.append(line)
+            line = line[1:-1]
+            fields = line.split('.')
+            words = fields[-2].split('_') + fields[-1].split('_')
+            max_num_words = max(len(words), max_num_words)
+            rel_words.append(words)
+            word_counter += words
+elif 'wikimovie' in dataset:
+    with open(rel_path) as f:
+        for line in tqdm(f.readlines()):
+            line = line.strip()
+            all_relations.append(line)
+            words = line.split('_')
+            max_num_words = max(len(words), max_num_words)
+            rel_words.append(words)
+            word_counter += words
+
+print('max_num_words: ', max_num_words)
+
+word_counter = nlp.data.count_tokens(word_counter)
+glove_emb = nlp.embedding.create('glove', source='glove.6B.100d')
+vocab = nlp.Vocab(word_counter)
+vocab.set_embedding(glove_emb)
+
+emb_mat = vocab.embedding.idx_to_vec.asnumpy()
+np.save(dataset + '/glove_word_emb_100d', emb_mat)
+
+with open(dataset + '/glove_vocab.txt', 'w') as g:
+    g.write('\n'.join(vocab.idx_to_token))
+
+assert False
+
+rel_word_ids = np.ones((len(rel_words) + 1, max_num_words), dtype=int) # leave the first 1 for padding relation
+rel_emb_mat = []
+for rel_idx, words in enumerate(rel_words):
+    for i, word in enumerate(words):
+        rel_word_ids[rel_idx + 1, i] = vocab.token_to_idx[word]
+
+np.save(dataset + '/rel_word_idx', rel_word_ids)
+
+all_relations = ['pad_rel'] + all_relations
+with open(rel_path, 'w') as g:
+    g.write('\n'.join(all_relations))
+
+
+