Update seq2seq

seq2seq/reader.py

@@ -16,203 +16,333 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import collections
-import os
+import glob
 import io
-import sys
 import numpy as np
-
-Py3 = sys.version_info[0] == 3
-
-UNK_ID = 0
-
-
-def _read_words(filename):
-    data = []
-    with io.open(filename, "r", encoding='utf-8') as f:
-        if Py3:
-            return f.read().replace("\n", "<eos>").split()
+import itertools
+from paddle.fluid.dygraph.parallel import ParallelEnv
+from paddle.fluid.io import BatchSampler, DataLoader, Dataset
+
+
+def prepare_train_input(insts, bos_id, eos_id, pad_id):
+    src, src_length = pad_batch_data(
+        [inst[0] for inst in insts], pad_id=pad_id)
+    trg, trg_length = pad_batch_data(
+        [[bos_id] + inst[1] + [eos_id] for inst in insts], pad_id=pad_id)
+    trg_length = trg_length - 1
+    return src, src_length, trg[:, :-1], trg_length, trg[:, 1:, np.newaxis]
+
+
+def pad_batch_data(insts, pad_id):
+    """
+    Pad the instances to the max sequence length in batch, and generate the
+    corresponding position data and attention bias.
+    """
+    inst_lens = np.array([len(inst) for inst in insts], dtype="int64")
+    max_len = np.max(inst_lens)
+    inst_data = np.array(
+        [inst + [pad_id] * (max_len - len(inst)) for inst in insts],
+        dtype="int64")
+    return inst_data, inst_lens
+
+
+class SortType(object):
+    GLOBAL = 'global'
+    POOL = 'pool'
+    NONE = "none"
+
+
+class Converter(object):
+    def __init__(self, vocab, beg, end, unk, delimiter, add_beg, add_end):
+        self._vocab = vocab
+        self._beg = beg
+        self._end = end
+        self._unk = unk
+        self._delimiter = delimiter
+        self._add_beg = add_beg
+        self._add_end = add_end
+
+    def __call__(self, sentence):
+        return ([self._beg] if self._add_beg else []) + [
+            self._vocab.get(w, self._unk)
+            for w in sentence.split(self._delimiter)
+        ] + ([self._end] if self._add_end else [])
+
+
+class ComposedConverter(object):
+    def __init__(self, converters):
+        self._converters = converters
+
+    def __call__(self, fields):
+        return [
+            converter(field)
+            for field, converter in zip(fields, self._converters)
+        ]
+
+
+class SentenceBatchCreator(object):
+    def __init__(self, batch_size):
+        self.batch = []
+        self._batch_size = batch_size
+
+    def append(self, info):
+        self.batch.append(info)
+        if len(self.batch) == self._batch_size:
+            tmp = self.batch
+            self.batch = []
+            return tmp
+
+
+class TokenBatchCreator(object):
+    def __init__(self, batch_size):
+        self.batch = []
+        self.max_len = -1
+        self._batch_size = batch_size
+
+    def append(self, info):
+        cur_len = info.max_len
+        max_len = max(self.max_len, cur_len)
+        if max_len * (len(self.batch) + 1) > self._batch_size:
+            result = self.batch
+            self.batch = [info]
+            self.max_len = cur_len
+            return result
         else:
-            return f.read().decode("utf-8").replace(u"\n", u"<eos>").split()
-
-
-def read_all_line(filenam):
-    data = []
-    with io.open(filename, "r", encoding='utf-8') as f:
-        for line in f.readlines():
-            data.append(line.strip())
-
-
-def _build_vocab(filename):
-
-    vocab_dict = {}
-    ids = 0
-    with io.open(filename, "r", encoding='utf-8') as f:
-        for line in f.readlines():
-            vocab_dict[line.strip()] = ids
-            ids += 1
-
-    print("vocab word num", ids)
-
-    return vocab_dict
-
-
-def _para_file_to_ids(src_file, tar_file, src_vocab, tar_vocab):
-
-    src_data = []
-    with io.open(src_file, "r", encoding='utf-8') as f_src:
-        for line in f_src.readlines():
-            arra = line.strip().split()
-            ids = [src_vocab[w] if w in src_vocab else UNK_ID for w in arra]
-            ids = ids
-
-            src_data.append(ids)
-
-    tar_data = []
-    with io.open(tar_file, "r", encoding='utf-8') as f_tar:
-        for line in f_tar.readlines():
-            arra = line.strip().split()
-            ids = [tar_vocab[w] if w in tar_vocab else UNK_ID for w in arra]
-
-            ids = [1] + ids + [2]
-
-            tar_data.append(ids)
-
-    return src_data, tar_data
-
-
-def filter_len(src, tar, max_sequence_len=50):
-    new_src = []
-    new_tar = []
-
-    for id1, id2 in zip(src, tar):
-        if len(id1) > max_sequence_len:
-            id1 = id1[:max_sequence_len]
-        if len(id2) > max_sequence_len + 2:
-            id2 = id2[:max_sequence_len + 2]
+            self.max_len = max_len
+            self.batch.append(info)
 
-        new_src.append(id1)
-        new_tar.append(id2)
 
-    return new_src, new_tar
+class SampleInfo(object):
+    def __init__(self, i, max_len, min_len):
+        self.i = i
+        self.min_len = min_len
+        self.max_len = max_len
 
 
-def raw_data(src_lang,
-             tar_lang,
-             vocab_prefix,
-             train_prefix,
-             eval_prefix,
-             test_prefix,
-             max_sequence_len=50):
+class MinMaxFilter(object):
+    def __init__(self, max_len, min_len, underlying_creator):
+        self._min_len = min_len
+        self._max_len = max_len
+        self._creator = underlying_creator
 
-    src_vocab_file = vocab_prefix + "." + src_lang
-    tar_vocab_file = vocab_prefix + "." + tar_lang
-
-    src_train_file = train_prefix + "." + src_lang
-    tar_train_file = train_prefix + "." + tar_lang
-
-    src_eval_file = eval_prefix + "." + src_lang
-    tar_eval_file = eval_prefix + "." + tar_lang
-
-    src_test_file = test_prefix + "." + src_lang
-    tar_test_file = test_prefix + "." + tar_lang
-
-    src_vocab = _build_vocab(src_vocab_file)
-    tar_vocab = _build_vocab(tar_vocab_file)
-
-    train_src, train_tar = _para_file_to_ids( src_train_file, tar_train_file, \
-                                              src_vocab, tar_vocab )
-    train_src, train_tar = filter_len(
-        train_src, train_tar, max_sequence_len=max_sequence_len)
-    eval_src, eval_tar = _para_file_to_ids( src_eval_file, tar_eval_file, \
-                                              src_vocab, tar_vocab )
-
-    test_src, test_tar = _para_file_to_ids( src_test_file, tar_test_file, \
-                                              src_vocab, tar_vocab )
-
-    return ( train_src, train_tar), (eval_src, eval_tar), (test_src, test_tar),\
-            (src_vocab, tar_vocab)
-
-
-def raw_mono_data(vocab_file, file_path):
+    def append(self, info):
+        if info.max_len > self._max_len or info.min_len < self._min_len:
+            return
+        else:
+            return self._creator.append(info)
+
+    @property
+    def batch(self):
+        return self._creator.batch
+
+
+class Seq2SeqDataset(Dataset):
+    def __init__(self,
+                 src_vocab_fpath,
+                 trg_vocab_fpath,
+                 fpattern,
+                 field_delimiter="\t",
+                 token_delimiter=" ",
+                 start_mark="<s>",
+                 end_mark="<e>",
+                 unk_mark="<unk>",
+                 only_src=False,
+                 trg_fpattern=None):
+        # convert str to bytes, and use byte data
+        # field_delimiter = field_delimiter.encode("utf8")
+        # token_delimiter = token_delimiter.encode("utf8")
+        # start_mark = start_mark.encode("utf8")
+        # end_mark = end_mark.encode("utf8")
+        # unk_mark = unk_mark.encode("utf8")
+        self._src_vocab = self.load_dict(src_vocab_fpath)
+        self._trg_vocab = self.load_dict(trg_vocab_fpath)
+        self._bos_idx = self._src_vocab[start_mark]
+        self._eos_idx = self._src_vocab[end_mark]
+        self._unk_idx = self._src_vocab[unk_mark]
+        self._only_src = only_src
+        self._field_delimiter = field_delimiter
+        self._token_delimiter = token_delimiter
+        self.load_src_trg_ids(fpattern, trg_fpattern)
+
+    def load_src_trg_ids(self, fpattern, trg_fpattern=None):
+        src_converter = Converter(
+            vocab=self._src_vocab,
+            beg=self._bos_idx,
+            end=self._eos_idx,
+            unk=self._unk_idx,
+            delimiter=self._token_delimiter,
+            add_beg=False,
+            add_end=False)
+
+        trg_converter = Converter(
+            vocab=self._trg_vocab,
+            beg=self._bos_idx,
+            end=self._eos_idx,
+            unk=self._unk_idx,
+            delimiter=self._token_delimiter,
+            add_beg=False,
+            add_end=False)
+
+        converters = ComposedConverter([src_converter, trg_converter])
+
+        self._src_seq_ids = []
+        self._trg_seq_ids = []
+        self._sample_infos = []
+
+        slots = [self._src_seq_ids, self._trg_seq_ids]
+        lens = []
+        for i, line in enumerate(self._load_lines(fpattern, trg_fpattern)):
+            lens = []
+            for field, slot in zip(converters(line), slots):
+                slot.append(field)
+                lens.append(len(field))
+            # self._sample_infos.append(SampleInfo(i, max(lens), min(lens)))
+            self._sample_infos.append(SampleInfo(i, lens[0], lens[0]))
+
+    def _load_lines(self, fpattern, trg_fpattern=None):
+        fpaths = glob.glob(fpattern)
+        fpaths = sorted(fpaths)  # TODO: Add custum sort
+        assert len(fpaths) > 0, "no matching file to the provided data path"
+
+        if trg_fpattern is None:
+            for fpath in fpaths:
+                # with io.open(fpath, "rb") as f:
+                with io.open(fpath, "r", encoding="utf8") as f:
+                    for line in f:
+                        fields = line.strip("\n").split(self._field_delimiter)
+                        yield fields
+        else:
+            # separated source and target language data files
+            # assume we can get aligned data by sort the two language files
+            # TODO: Need more rigorous check
+            trg_fpaths = glob.glob(trg_fpattern)
+            trg_fpaths = sorted(trg_fpaths)
+            assert len(fpaths) == len(
+                trg_fpaths
+            ), "the number of source language data files must equal \
+                with that of source language"
+
+            for fpath, trg_fpath in zip(fpaths, trg_fpaths):
+                # with io.open(fpath, "rb") as f:
+                #     with io.open(trg_fpath, "rb") as trg_f:
+                with io.open(fpath, "r", encoding="utf8") as f:
+                    with io.open(trg_fpath, "r", encoding="utf8") as trg_f:
+                        for line in zip(f, trg_f):
+                            fields = [field.strip("\n") for field in line]
+                            yield fields
+
+    @staticmethod
+    def load_dict(dict_path, reverse=False):
+        word_dict = {}
+        # with io.open(dict_path, "rb") as fdict:
+        with io.open(dict_path, "r", encoding="utf8") as fdict:
+            for idx, line in enumerate(fdict):
+                if reverse:
+                    word_dict[idx] = line.strip("\n")
+                else:
+                    word_dict[line.strip("\n")] = idx
+        return word_dict
 
-    src_vocab = _build_vocab(vocab_file)
+    def get_vocab_summary(self):
+        return len(self._src_vocab), len(
+            self._trg_vocab), self._bos_idx, self._eos_idx, self._unk_idx
 
-    test_src, test_tar = _para_file_to_ids( file_path, file_path, \
-                                              src_vocab, src_vocab )
+    def __getitem__(self, idx):
+        return (self._src_seq_ids[idx], self._trg_seq_ids[idx]
+                ) if self._trg_seq_ids else self._src_seq_ids[idx]
 
-    return (test_src, test_tar)
+    def __len__(self):
+        return len(self._sample_infos)
 
 
-def get_data_iter(raw_data,
+class Seq2SeqBatchSampler(BatchSampler):
+    def __init__(self,
+                 dataset,
                  batch_size,
-                  mode='train',
-                  enable_ce=False,
-                  cache_num=20):
-
-    src_data, tar_data = raw_data
-
-    data_len = len(src_data)
-
-    index = np.arange(data_len)
-    if mode == "train" and not enable_ce:
-        np.random.shuffle(index)
-
-    def to_pad_np(data, source=False):
-        max_len = 0
-        bs = min(batch_size, len(data))
-        for ele in data:
-            if len(ele) > max_len:
-                max_len = len(ele)
-
-        ids = np.ones((bs, max_len), dtype='int64') * 2
-        mask = np.zeros((bs), dtype='int32')
-
-        for i, ele in enumerate(data):
-            ids[i, :len(ele)] = ele
-            if not source:
-                mask[i] = len(ele) - 1
+                 pool_size=10000,
+                 sort_type=SortType.NONE,
+                 min_length=0,
+                 max_length=100,
+                 shuffle=False,
+                 shuffle_batch=False,
+                 use_token_batch=False,
+                 clip_last_batch=False,
+                 seed=None):
+        for arg, value in locals().items():
+            if arg != "self":
+                setattr(self, "_" + arg, value)
+        self._random = np.random
+        self._random.seed(seed)
+        # for multi-devices
+        self._nranks = ParallelEnv().nranks
+        self._local_rank = ParallelEnv().local_rank
+        self._device_id = ParallelEnv().dev_id
+
+    def __iter__(self):
+        # global sort or global shuffle
+        if self._sort_type == SortType.GLOBAL:
+            infos = sorted(
+                self._dataset._sample_infos, key=lambda x: x.max_len)
         else:
-                mask[i] = len(ele)
-
-        return ids, mask
-
-    b_src = []
-
-    if mode != "train":
-        cache_num = 1
-    for j in range(data_len):
-        if len(b_src) == batch_size * cache_num:
-            # build batch size
-
-            # sort
-            if mode == 'infer':
-                new_cache = b_src
+            if self._shuffle:
+                infos = self._dataset._sample_infos
+                self._random.shuffle(infos)
             else:
-                new_cache = sorted(b_src, key=lambda k: len(k[0]))
-
-            for i in range(cache_num):
-                batch_data = new_cache[i * batch_size:(i + 1) * batch_size]
-                src_cache = [w[0] for w in batch_data]
-                tar_cache = [w[1] for w in batch_data]
-                src_ids, src_mask = to_pad_np(src_cache, source=True)
-                tar_ids, tar_mask = to_pad_np(tar_cache)
-                yield (src_ids, src_mask, tar_ids, tar_mask)
-
-            b_src = []
-
-        b_src.append((src_data[index[j]], tar_data[index[j]]))
-    if len(b_src) == batch_size * cache_num or mode == 'infer':
-        if mode == 'infer':
-            new_cache = b_src
+                infos = self._dataset._sample_infos
+
+            if self._sort_type == SortType.POOL:
+                reverse = True
+                for i in range(0, len(infos), self._pool_size):
+                    # to avoid placing short next to long sentences
+                    reverse = not reverse
+                    infos[i:i + self._pool_size] = sorted(
+                        infos[i:i + self._pool_size],
+                        key=lambda x: x.max_len,
+                        reverse=reverse)
+
+        batches = []
+        batch_creator = TokenBatchCreator(
+            self.
+            _batch_size) if self._use_token_batch else SentenceBatchCreator(
+                self._batch_size * self._nranks)
+        batch_creator = MinMaxFilter(self._max_length, self._min_length,
+                                     batch_creator)
+
+        for info in infos:
+            batch = batch_creator.append(info)
+            if batch is not None:
+                batches.append(batch)
+
+        if not self._clip_last_batch and len(batch_creator.batch) != 0:
+            batches.append(batch_creator.batch)
+
+        if self._shuffle_batch:
+            self._random.shuffle(batches)
+
+        if not self._use_token_batch:
+            # when producing batches according to sequence number, to confirm
+            # neighbor batches which would be feed and run parallel have similar
+            # length (thus similar computational cost) after shuffle, we as take
+            # them as a whole when shuffling and split here
+            batches = [[
+                batch[self._batch_size * i:self._batch_size * (i + 1)]
+                for i in range(self._nranks)
+            ] for batch in batches]
+            batches = list(itertools.chain.from_iterable(batches))
+
+        # for multi-device
+        for batch_id, batch in enumerate(batches):
+            if batch_id % self._nranks == self._local_rank:
+                batch_indices = [info.i for info in batch]
+                yield batch_indices
+        if self._local_rank > len(batches) % self._nranks:
+            yield batch_indices
+
+    def __len__(self):
+        if not self._use_token_batch:
+            batch_number = (
+                len(self._dataset) + self._batch_size * self._nranks - 1) // (
+                    self._batch_size * self._nranks)
         else:
-            new_cache = sorted(b_src, key=lambda k: len(k[0]))
-
-        for i in range(cache_num):
-            batch_end = min(len(new_cache), (i + 1) * batch_size)
-            batch_data = new_cache[i * batch_size:batch_end]
-            src_cache = [w[0] for w in batch_data]
-            tar_cache = [w[1] for w in batch_data]
-            src_ids, src_mask = to_pad_np(src_cache, source=True)
-            tar_ids, tar_mask = to_pad_np(tar_cache)
-            yield (src_ids, src_mask, tar_ids, tar_mask)
+            batch_number = 100
+        return batch_number

seq2seq/seq2seq_attn.py

@@ -41,9 +41,10 @@ class AttentionLayer(Layer):
             bias_attr=bias)
 
     def forward(self, hidden, encoder_output, encoder_padding_mask):
-        query = self.input_proj(hidden)
+        # query = self.input_proj(hidden)
+        encoder_output = self.input_proj(encoder_output)
         attn_scores = layers.matmul(
-            layers.unsqueeze(query, [1]), encoder_output, transpose_y=True)
+            layers.unsqueeze(hidden, [1]), encoder_output, transpose_y=True)
         if encoder_padding_mask is not None:
             attn_scores = layers.elementwise_add(attn_scores,
                                                  encoder_padding_mask)
@@ -73,7 +74,9 @@ class DecoderCell(RNNCell):
                     BasicLSTMCell(
                         input_size=input_size + hidden_size
                         if i == 0 else hidden_size,
-                        hidden_size=hidden_size)))
+                        hidden_size=hidden_size,
+                        param_attr=ParamAttr(initializer=UniformInitializer(
+                            low=-init_scale, high=init_scale)))))
         self.attention_layer = AttentionLayer(hidden_size)
 
     def forward(self,
@@ -107,8 +110,8 @@ class Decoder(Layer):
             size=[vocab_size, embed_dim],
             param_attr=ParamAttr(initializer=UniformInitializer(
                 low=-init_scale, high=init_scale)))
-        self.lstm_attention = RNN(DecoderCell(num_layers, embed_dim,
-                                              hidden_size, init_scale),
+        self.lstm_attention = RNN(DecoderCell(
+            num_layers, embed_dim, hidden_size, dropout_prob, init_scale),
                                   is_reverse=False,
                                   time_major=False)
         self.output_layer = Linear(

seq2seq/seq2seq_base.py

@@ -86,7 +86,7 @@ class Encoder(Layer):
             param_attr=ParamAttr(initializer=UniformInitializer(
                 low=-init_scale, high=init_scale)))
         self.stack_lstm = RNN(EncoderCell(num_layers, embed_dim, hidden_size,
-                                          init_scale),
+                                          dropout_prob, init_scale),
                               is_reverse=False,
                               time_major=False)
 
@@ -114,7 +114,7 @@ class Decoder(Layer):
             param_attr=ParamAttr(initializer=UniformInitializer(
                 low=-init_scale, high=init_scale)))
         self.stack_lstm = RNN(DecoderCell(num_layers, embed_dim, hidden_size,
-                                          init_scale),
+                                          dropout_prob, init_scale),
                               is_reverse=False,
                               time_major=False)
         self.output_layer = Linear(

seq2seq/train.py

@@ -17,8 +17,7 @@ import os
 import six
 import sys
 sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
-import time
-import contextlib
+import random
 from functools import partial
 
 import numpy as np
@@ -34,16 +33,17 @@ from seq2seq_base import BaseModel, CrossEntropyCriterion
 from seq2seq_attn import AttentionModel
 from model import Input, set_device
 from callbacks import ProgBarLogger
-from metrics import Metric
-
-
-class PPL(Metric):
-    pass
+from reader import Seq2SeqDataset, Seq2SeqBatchSampler, SortType, prepare_train_input
 
 
 def do_train(args):
     device = set_device("gpu" if args.use_gpu else "cpu")
-    fluid.enable_dygraph(device)  #if args.eager_run else None
+    fluid.enable_dygraph(device) if args.eager_run else None
+
+    if args.enable_ce:
+        fluid.default_main_program().random_seed = 102
+        fluid.default_startup_program().random_seed = 102
+        args.shuffle = False
 
     # define model
     inputs = [
@@ -58,6 +58,45 @@ def do_train(args):
     ]
     labels = [Input([None, None, 1], "int64", name="label"), ]
 
+    # def dataloader
+    data_loaders = [None, None]
+    data_prefixes = [args.train_data_prefix, args.eval_data_prefix
+                     ] if args.eval_data_prefix else [args.train_data_prefix]
+    for i, data_prefix in enumerate(data_prefixes):
+        dataset = Seq2SeqDataset(
+            fpattern=data_prefix + "." + args.src_lang,
+            trg_fpattern=data_prefix + "." + args.tar_lang,
+            src_vocab_fpath=args.vocab_prefix + "." + args.src_lang,
+            trg_vocab_fpath=args.vocab_prefix + "." + args.tar_lang,
+            token_delimiter=None,
+            start_mark="<s>",
+            end_mark="</s>",
+            unk_mark="<unk>")
+        (args.src_vocab_size, args.trg_vocab_size, bos_id, eos_id,
+         unk_id) = dataset.get_vocab_summary()
+        batch_sampler = Seq2SeqBatchSampler(
+            dataset=dataset,
+            use_token_batch=False,
+            batch_size=args.batch_size,
+            pool_size=args.batch_size * 20,
+            sort_type=SortType.POOL,
+            shuffle=args.shuffle)
+        data_loader = DataLoader(
+            dataset=dataset,
+            batch_sampler=batch_sampler,
+            places=device,
+            feed_list=None if fluid.in_dygraph_mode() else
+            [x.forward() for x in inputs + labels],
+            collate_fn=partial(
+                prepare_train_input,
+                bos_id=bos_id,
+                eos_id=eos_id,
+                pad_id=eos_id),
+            num_workers=0,
+            return_list=True)
+        data_loaders[i] = data_loader
+    train_loader, eval_loader = data_loaders
+
     model = AttentionModel(args.src_vocab_size, args.tar_vocab_size,
                            args.hidden_size, args.hidden_size, args.num_layers,
                            args.dropout)
@@ -69,39 +108,12 @@ def do_train(args):
         CrossEntropyCriterion(),
         inputs=inputs,
         labels=labels)
-
-    batch_size = 32
-    src_seq_len = 10
-    trg_seq_len = 12
-    iter_num = 10
-
-    def random_generator():
-        for i in range(iter_num):
-            src = np.random.randint(2, args.src_vocab_size,
-                                    (batch_size, src_seq_len)).astype("int64")
-            src_length = np.random.randint(1, src_seq_len,
-                                           (batch_size, )).astype("int64")
-            trg = np.random.randint(2, args.tar_vocab_size,
-                                    (batch_size, trg_seq_len)).astype("int64")
-            trg_length = np.random.randint(1, trg_seq_len,
-                                           (batch_size, )).astype("int64")
-            label = np.random.randint(
-                1, trg_seq_len, (batch_size, trg_seq_len, 1)).astype("int64")
-            yield src, src_length, trg, trg_length, label
-
-    model.fit(train_data=random_generator, log_freq=1)
-    exit(0)
-
-    data_loaders = [None, None]
-    data_files = [args.training_file, args.validation_file
-                  ] if args.validation_file else [args.training_file]
-    train_loader, eval_loader = data_loaders
-
     model.fit(train_data=train_loader,
-              eval_data=None,
+              eval_data=eval_loader,
               epochs=1,
               eval_freq=1,
               save_freq=1,
+              log_freq=1,
               verbose=2)