bug fix for w2v

core/trainers/framework/dataset.py

@@ -68,6 +68,8 @@ class DataLoader(DatasetBase):
             reader_ins = SlotReader(context["config_yaml"])
         if hasattr(reader_ins, 'generate_batch_from_trainfiles'):
             dataloader.set_sample_list_generator(reader)
+        elif hasattr(reader_ins, 'batch_tensor_creator'):
+            dataloader.set_batch_generator(reader)
         else:
             dataloader.set_sample_generator(reader, batch_size)
         return dataloader

core/utils/dataloader_instance.py

@@ -67,6 +67,10 @@ def dataloader_by_name(readerclass,
 
     if hasattr(reader, 'generate_batch_from_trainfiles'):
         return gen_batch_reader()
+
+    if hasattr(reader, "batch_tensor_creator"):
+        return reader.batch_tensor_creator(gen_reader)
+
     return gen_reader
 
 

models/recall/word2vec/README.md

@@ -19,6 +19,8 @@
 ├── data_prepare.sh #一键数据处理脚本
 ├── w2v_reader.py #训练数据reader
 ├── w2v_evaluate_reader.py # 预测数据reader
+├── infer.py # 自定义预测脚本
+├── utils.py # 自定义预测中用到的reader等工具
 ```
 
 注：在阅读该示例前，建议您先了解以下内容：
@@ -154,9 +156,12 @@ runner:
   phases: [phase1]
 ```
 ### 单机预测
+我们通过词类比（Word Analogy）任务来检验word2vec模型的训练效果。输入四个词A，B，C，D，假设存在一种关系relation, 使得relation(A, B) = relation(C， D)，然后通过A，B，C去预测D，emb(D) = emb(B) - emb(A) + emb(C)。
 
 CPU环境
 
+PaddleRec预测配置：
+
 在config.yaml文件中设置好epochs、device等参数。
 
 ```
@@ -168,6 +173,10 @@ CPU环境
   print_interval: 1
   phases: [phase2]
 ```
+为复现论文效果，我们提供了一个自定义预测脚本，自定义预测中，我们会跳过预测结果是输入A，B，C的情况，计算预测准确率。执行命令如下：
+```
+python infer.py --test_dir ./data/test --dict_path ./data/dict/word_id_dict.txt --batch_size 20000 --model_dir ./increment_w2v/  --start_index 0 --last_index 5 --emb_size 300
+```
 
 ### 运行
 ```
@@ -212,13 +221,12 @@ Infer phase2 of epoch 3 done, use time: 4.43099021912, global metrics: acc=[1.]
 - batch_size: 修改config.yaml中dataset_train数据集的batch_size为100。
 - epochs: 修改config.yaml中runner的epochs为5。
 
-使用cpu训练 5轮 测试Recall@20:0.540
-
 修改后运行方案：修改config.yaml中的'workspace'为config.yaml的目录位置，执行
 ```
 python -m paddlerec.run -m /home/your/dir/config.yaml #调试模式 直接指定本地config的绝对路径
 ```
 
+使用cpu训练5轮，自定义测试（跳过输入）准确率为0.540。
 ## 进阶使用
 
 ## FAQ

models/recall/word2vec/config.yaml

@@ -22,7 +22,7 @@ dataset:
   word_count_dict_path: "{workspace}/data/dict/word_count_dict.txt"
   data_converter: "{workspace}/w2v_reader.py"
 - name: dataset_infer # name
-  batch_size: 50
+  batch_size: 2000
   type: DataLoader # or QueueDataset
   data_path: "{workspace}/data/test"
   word_id_dict_path: "{workspace}/data/dict/word_id_dict.txt"
@@ -59,7 +59,7 @@ runner:
   save_inference_feed_varnames: [] # feed vars of save inference
   save_inference_fetch_varnames: [] # fetch vars of save inference
   init_model_path: "" # load model path
-  print_interval: 1
+  print_interval: 1000
   phases: [phase1]
 - name: single_cpu_infer
   class: infer

models/recall/word2vec/data_prepare.sh

@@ -25,7 +25,7 @@ mv 1-billion-word-language-modeling-benchmark-r13output/training-monolingual.tok
 python preprocess.py --build_dict --build_dict_corpus_dir raw_data/training-monolingual.tokenized.shuffled --dict_path raw_data/word_count_dict.txt
 python preprocess.py --filter_corpus --dict_path raw_data/word_count_dict.txt --input_corpus_dir raw_data/training-monolingual.tokenized.shuffled --output_corpus_dir raw_data/convert_text8 --min_count 5 --downsample 0.001
 mv raw_data/word_count_dict.txt data/dict/
-mv raw_data/word_id_dict.txt data/dict/
+mv raw_data/word_count_dict.txt_word_to_id_ data/dict/word_id_dict.txt
 
 rm -rf data/train/*
 rm -rf data/test/*

models/recall/word2vec/infer.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import sys
+import time
+import math
+import numpy as np
+import six
+import paddle.fluid as fluid
+import paddle
+import utils
+if six.PY2:
+    reload(sys)
+    sys.setdefaultencoding('utf-8')
+
+
+def parse_args():
+    parser = argparse.ArgumentParser("PaddlePaddle Word2vec infer example")
+    parser.add_argument(
+        '--dict_path',
+        type=str,
+        default='./data/data_c/1-billion_dict_word_to_id_',
+        help="The path of dic")
+    parser.add_argument(
+        '--test_dir', type=str, default='test_data', help='test file address')
+    parser.add_argument(
+        '--print_step', type=int, default='500000', help='print step')
+    parser.add_argument(
+        '--start_index', type=int, default='0', help='start index')
+    parser.add_argument(
+        '--last_index', type=int, default='100', help='last index')
+    parser.add_argument(
+        '--model_dir', type=str, default='model', help='model dir')
+    parser.add_argument(
+        '--use_cuda', type=int, default='0', help='whether use cuda')
+    parser.add_argument(
+        '--batch_size', type=int, default='5', help='batch_size')
+    parser.add_argument(
+        '--emb_size', type=int, default='64', help='batch_size')
+    args = parser.parse_args()
+    return args
+
+
+def infer_network(vocab_size, emb_size):
+    analogy_a = fluid.data(name="analogy_a", shape=[None], dtype='int64')
+    analogy_b = fluid.data(name="analogy_b", shape=[None], dtype='int64')
+    analogy_c = fluid.data(name="analogy_c", shape=[None], dtype='int64')
+    all_label = fluid.data(name="all_label", shape=[vocab_size], dtype='int64')
+    emb_all_label = fluid.embedding(
+        input=all_label, size=[vocab_size, emb_size], param_attr="emb")
+
+    emb_a = fluid.embedding(
+        input=analogy_a, size=[vocab_size, emb_size], param_attr="emb")
+    emb_b = fluid.embedding(
+        input=analogy_b, size=[vocab_size, emb_size], param_attr="emb")
+    emb_c = fluid.embedding(
+        input=analogy_c, size=[vocab_size, emb_size], param_attr="emb")
+    target = fluid.layers.elementwise_add(
+        fluid.layers.elementwise_sub(emb_b, emb_a), emb_c)
+    emb_all_label_l2 = fluid.layers.l2_normalize(x=emb_all_label, axis=1)
+    dist = fluid.layers.matmul(x=target, y=emb_all_label_l2, transpose_y=True)
+    values, pred_idx = fluid.layers.topk(input=dist, k=4)
+    return values, pred_idx
+
+
+def infer_epoch(args, vocab_size, test_reader, use_cuda, i2w):
+    """ inference function """
+    place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace()
+    exe = fluid.Executor(place)
+    emb_size = args.emb_size
+    batch_size = args.batch_size
+    with fluid.scope_guard(fluid.Scope()):
+        main_program = fluid.Program()
+        with fluid.program_guard(main_program):
+            values, pred = infer_network(vocab_size, emb_size)
+            for epoch in range(start_index, last_index + 1):
+                copy_program = main_program.clone()
+                model_path = model_dir + "/" + str(epoch)
+                fluid.io.load_persistables(
+                    exe, model_path, main_program=copy_program)
+                accum_num = 0
+                accum_num_sum = 0.0
+                t0 = time.time()
+                step_id = 0
+                for data in test_reader():
+                    step_id += 1
+                    b_size = len([dat[0] for dat in data])
+                    wa = np.array([dat[0] for dat in data]).astype(
+                        "int64").reshape(b_size)
+                    wb = np.array([dat[1] for dat in data]).astype(
+                        "int64").reshape(b_size)
+                    wc = np.array([dat[2] for dat in data]).astype(
+                        "int64").reshape(b_size)
+
+                    label = [dat[3] for dat in data]
+                    input_word = [dat[4] for dat in data]
+                    para = exe.run(copy_program,
+                                   feed={
+                                       "analogy_a": wa,
+                                       "analogy_b": wb,
+                                       "analogy_c": wc,
+                                       "all_label": np.arange(vocab_size)
+                                       .reshape(vocab_size).astype("int64"),
+                                   },
+                                   fetch_list=[pred.name, values],
+                                   return_numpy=False)
+                    pre = np.array(para[0])
+                    val = np.array(para[1])
+                    for ii in range(len(label)):
+                        top4 = pre[ii]
+                        accum_num_sum += 1
+                        for idx in top4:
+                            if int(idx) in input_word[ii]:
+                                continue
+                            if int(idx) == int(label[ii][0]):
+                                accum_num += 1
+                            break
+                    if step_id % 1 == 0:
+                        print("step:%d %d " % (step_id, accum_num))
+
+                print("epoch:%d \t acc:%.3f " %
+                      (epoch, 1.0 * accum_num / accum_num_sum))
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    start_index = args.start_index
+    last_index = args.last_index
+    test_dir = args.test_dir
+    model_dir = args.model_dir
+    batch_size = args.batch_size
+    dict_path = args.dict_path
+    use_cuda = True if args.use_cuda else False
+    print("start index: ", start_index, " last_index:", last_index)
+    vocab_size, test_reader, id2word = utils.prepare_data(
+        test_dir, dict_path, batch_size=batch_size)
+    print("vocab_size:", vocab_size)
+    infer_epoch(
+        args,
+        vocab_size,
+        test_reader=test_reader,
+        use_cuda=use_cuda,
+        i2w=id2word)

models/recall/word2vec/model.py

@@ -209,10 +209,10 @@ class Model(ModelBase):
         emb_all_label_l2 = fluid.layers.l2_normalize(x=emb_all_label, axis=1)
         dist = fluid.layers.matmul(
             x=target, y=emb_all_label_l2, transpose_y=True)
-        values, pred_idx = fluid.layers.topk(input=dist, k=4)
+        values, pred_idx = fluid.layers.topk(input=dist, 1)
         label = fluid.layers.expand(
             fluid.layers.unsqueeze(
-                inputs[3], axes=[1]), expand_times=[1, 4])
+                inputs[3], axes=[1]), expand_times=[1, 1])
         label_ones = fluid.layers.fill_constant_batch_size_like(
             label, shape=[-1, 1], value=1.0, dtype='float32')
         right_cnt = fluid.layers.reduce_sum(input=fluid.layers.cast(

models/recall/word2vec/utils.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import sys
+import collections
+import six
+import time
+import numpy as np
+import paddle.fluid as fluid
+import paddle
+import os
+import preprocess
+import io
+
+
+def BuildWord_IdMap(dict_path):
+    word_to_id = dict()
+    id_to_word = dict()
+    with io.open(dict_path, 'r', encoding='utf-8') as f:
+        for line in f:
+            word_to_id[line.split(' ')[0]] = int(line.split(' ')[1])
+            id_to_word[int(line.split(' ')[1])] = line.split(' ')[0]
+    return word_to_id, id_to_word
+
+
+def prepare_data(file_dir, dict_path, batch_size):
+    w2i, i2w = BuildWord_IdMap(dict_path)
+    vocab_size = len(i2w)
+    reader = fluid.io.batch(test(file_dir, w2i), batch_size)
+    return vocab_size, reader, i2w
+
+
+def check_version(with_shuffle_batch=False):
+    """
+     Log error and exit when the installed version of paddlepaddle is
+     not satisfied.
+     """
+    err = "PaddlePaddle version 1.6 or higher is required, " \
+          "or a suitable develop version is satisfied as well. \n" \
+          "Please make sure the version is good with your code." \
+
+    try:
+        if with_shuffle_batch:
+            fluid.require_version('1.7.0')
+        else:
+            fluid.require_version('1.6.0')
+    except Exception as e:
+        logger.error(err)
+        sys.exit(1)
+
+
+def native_to_unicode(s):
+    if _is_unicode(s):
+        return s
+    try:
+        return _to_unicode(s)
+    except UnicodeDecodeError:
+        res = _to_unicode(s, ignore_errors=True)
+        return res
+
+
+def _is_unicode(s):
+    if six.PY2:
+        if isinstance(s, unicode):
+            return True
+    else:
+        if isinstance(s, str):
+            return True
+    return False
+
+
+def _to_unicode(s, ignore_errors=False):
+    if _is_unicode(s):
+        return s
+    error_mode = "ignore" if ignore_errors else "strict"
+    return s.decode("utf-8", errors=error_mode)
+
+
+def strip_lines(line, vocab):
+    return _replace_oov(vocab, native_to_unicode(line))
+
+
+def _replace_oov(original_vocab, line):
+    """Replace out-of-vocab words with "<UNK>".
+  This maintains compatibility with published results.
+  Args:
+    original_vocab: a set of strings (The standard vocabulary for the dataset)
+    line: a unicode string - a space-delimited sequence of words.
+  Returns:
+    a unicode string - a space-delimited sequence of words.
+  """
+    return u" ".join([
+        word if word in original_vocab else u"<UNK>" for word in line.split()
+    ])
+
+
+def reader_creator(file_dir, word_to_id):
+    def reader():
+        files = os.listdir(file_dir)
+        for fi in files:
+            with io.open(
+                    os.path.join(file_dir, fi), "r", encoding='utf-8') as f:
+                for line in f:
+                    if ':' in line:
+                        pass
+                    else:
+                        line = strip_lines(line.lower(), word_to_id)
+                        line = line.split()
+                        yield [word_to_id[line[0]]], [word_to_id[line[1]]], [
+                            word_to_id[line[2]]
+                        ], [word_to_id[line[3]]], [
+                            word_to_id[line[0]], word_to_id[line[1]],
+                            word_to_id[line[2]]
+                        ]
+
+    return reader
+
+
+def test(test_dir, w2i):
+    return reader_creator(test_dir, w2i)

models/recall/word2vec/w2v_evaluate_reader.py

@@ -76,7 +76,7 @@ class Reader(ReaderBase):
     def generate_sample(self, line):
         def reader():
             if ':' in line:
-                pass
+                return
             features = self.strip_lines(line.lower(), self.word_to_id)
             features = features.split()
             yield [('analogy_a', [self.word_to_id[features[0]]]),

models/recall/word2vec/w2v_reader.py

@@ -15,6 +15,7 @@
 import io
 
 import numpy as np
+import paddle.fluid as fluid
 
 from paddlerec.core.reader import ReaderBase
 from paddlerec.core.utils import envs
@@ -47,6 +48,10 @@ class Reader(ReaderBase):
         self.with_shuffle_batch = envs.get_global_env(
             "hyper_parameters.with_shuffle_batch")
         self.random_generator = NumpyRandomInt(1, self.window_size + 1)
+        self.batch_size = envs.get_global_env(
+            "dataset.dataset_train.batch_size")
+        self.is_dataloader = envs.get_global_env(
+            "dataset.dataset_train.type") == "DataLoader"
 
         self.cs = None
         if not self.with_shuffle_batch:
@@ -88,7 +93,9 @@ class Reader(ReaderBase):
                 for context_id in context_word_ids:
                     output = [('input_word', [int(target_id)]),
                               ('true_label', [int(context_id)])]
-                    if not self.with_shuffle_batch:
+                    if self.with_shuffle_batch or self.is_dataloader:
+                        yield output
+                    else:
                         neg_array = self.cs.searchsorted(
                             np.random.sample(self.neg_num))
                         output += [('neg_label',
@@ -96,3 +103,36 @@ class Reader(ReaderBase):
                         yield output
 
         return reader
+
+    def batch_tensor_creator(self, sample_reader):
+        def __reader__():
+            result = [[], []]
+            for sample in sample_reader():
+                for i, fea in enumerate(sample):
+                    result[i].append(fea)
+                if len(result[0]) == self.batch_size:
+                    tensor_result = []
+                    for tensor in result:
+                        t = fluid.Tensor()
+                        dat = np.array(tensor, dtype='int64')
+                        if len(dat.shape) > 2:
+                            dat = dat.reshape((dat.shape[0], dat.shape[2]))
+                        elif len(dat.shape) == 1:
+                            dat = dat.reshape((-1, 1))
+                        t.set(dat, fluid.CPUPlace())
+                        tensor_result.append(t)
+                    if self.with_shuffle_batch:
+                        yield tensor_result
+                    else:
+                        tt = fluid.Tensor()
+                        neg_array = self.cs.searchsorted(
+                            np.random.sample(self.neg_num))
+                        neg_array = np.tile(neg_array, self.batch_size)
+                        tt.set(
+                            neg_array.reshape((self.batch_size, self.neg_num)),
+                            fluid.CPUPlace())
+                        tensor_result.append(tt)
+                        yield tensor_result
+                    result = [[], []]
+
+        return __reader__