zhpmatrix / Bertem
Licence: apache-2.0
论文实现(ACL2019):《Matching the Blanks: Distributional Similarity for Relation Learning》
Stars: ✭ 146
Projects that are alternatives of or similar to Bertem
Cnn Re Tf
Convolutional Neural Network for Multi-label Multi-instance Relation Extraction in Tensorflow
Stars: ✭ 190 (+30.14%)
Mutual labels: jupyter-notebook, relation-extraction
Ruijin round2
瑞金医院MMC人工智能辅助构建知识图谱大赛复赛
Stars: ✭ 159 (+8.9%)
Mutual labels: jupyter-notebook, relation-extraction
Pytorch graph Rel
A PyTorch implementation of GraphRel
Stars: ✭ 204 (+39.73%)
Mutual labels: jupyter-notebook, relation-extraction
Relation Classification Using Bidirectional Lstm Tree
TensorFlow Implementation of the paper "End-to-End Relation Extraction using LSTMs on Sequences and Tree Structures" and "Classifying Relations via Long Short Term Memory Networks along Shortest Dependency Paths" for classifying relations
Stars: ✭ 167 (+14.38%)
Mutual labels: jupyter-notebook, relation-extraction
Shufflenet V2 Tensorflow
A lightweight convolutional neural network
Stars: ✭ 145 (-0.68%)
Mutual labels: jupyter-notebook
Deep Deep
Adaptive crawler which uses Reinforcement Learning methods
Stars: ✭ 145 (-0.68%)
Mutual labels: jupyter-notebook
Data Driven Prediction Of Battery Cycle Life Before Capacity Degradation
Code for Nature energy manuscript
Stars: ✭ 145 (-0.68%)
Mutual labels: jupyter-notebook
Scipy con 2019
Tutorial Sessions for SciPy Con 2019
Stars: ✭ 142 (-2.74%)
Mutual labels: jupyter-notebook
100daysofmlcode
My journey to learn and grow in the domain of Machine Learning and Artificial Intelligence by performing the #100DaysofMLCode Challenge.
Stars: ✭ 146 (+0%)
Mutual labels: jupyter-notebook
Python Machine Learning Book
The "Python Machine Learning (1st edition)" book code repository and info resource
Stars: ✭ 11,428 (+7727.4%)
Mutual labels: jupyter-notebook
Hypertools Paper Notebooks
Supporting notebooks and data from hypertools paper
Stars: ✭ 145 (-0.68%)
Mutual labels: jupyter-notebook
Textbook
Principles and Techniques of Data Science, the textbook for Data 100 at UC Berkeley
Stars: ✭ 145 (-0.68%)
Mutual labels: jupyter-notebook
Rloss
Regularized Losses (rloss) for Weakly-supervised CNN Segmentation
Stars: ✭ 145 (-0.68%)
Mutual labels: jupyter-notebook
Deep Learning With Tensorflow Book
深度学习入门开源书,基于TensorFlow 2.0案例实战。Open source Deep Learning book, based on TensorFlow 2.0 framework.
Stars: ✭ 12,105 (+8191.1%)
Mutual labels: jupyter-notebook
Machinelearning Az
Repositorio del Curso de Machine Learning de la A a la Z con R y Python
Stars: ✭ 144 (-1.37%)
Mutual labels: jupyter-notebook
Citeomatic
A citation recommendation system that allows users to find relevant citations for their paper drafts. The tool is backed by Semantic Scholar's OpenCorpus dataset.
Stars: ✭ 145 (-0.68%)
Mutual labels: jupyter-notebook
Digital video introduction
A hands-on introduction to video technology: image, video, codec (av1, vp9, h265) and more (ffmpeg encoding).
Stars: ✭ 12,184 (+8245.21%)
Mutual labels: jupyter-notebook
Ppdai risk evaluation
“魔镜杯”风控算法大赛 拍拍贷风控模型,接近冠军分数
Stars: ✭ 144 (-1.37%)
Mutual labels: jupyter-notebook
实现说明
主要实现文章前半部分的工作,PyTorch实现,基于huggingface的工作,PyTorch才是世界上最屌的框架,逃。
实现参考
代码说明
(1)主要修改:modeling.py
output representation: BertForSequenceClassification
input representation: BertEmbeddings
input和output都实现了多种策略,可以结合具体的任务,找到最佳的组合。
(2)非主要实现:examples下的关于classification的文件
(3)服务部署:基于Flask,可以在本地开启一个服务。具体实现在tacred_run_infer.py中。
(4)代码仅供参考,不提供数据集,不提供预训练模型,不提供训练后的模型(希望理解吧)。
(5)相关工作可以参考我的博客-神经关系抽取,可能比这个代码更有价值一些吧。
实现结果:
数据集TACRED上的结果:
模型序号 | 输入类型 | 输出类型 | 指标类型 | P | R | F1 | 备注 |
---|---|---|---|---|---|---|---|
0 | entity marker | sum(entity start) | micro | 0.68 | 0.63 | 0.65 | base-model,lr=3e-5,epoch=3 |
macro | 0.60 | 0.54 | 0.55 | ||||
1 | entity marker | sum(entity start) | micro | 0.70 | 0.62 | 0.65 | large-model,lr=3e-5,epoch=1 |
macro | 0.63 | 0.52 | 0.55 | ||||
-1 | None | None | micro | 0.69 | 0.66 | 0.67 | 手误之后,再也找不到了,尴尬 |
macro | 0.58 | 0.50 | 0.53 |
数据集SemEval2010 Task 8上的结果:
模型序号 | 输入类型 | 输出类型 | 指标类型 | P | R | F1 | 备注 |
---|---|---|---|---|---|---|---|
0 | entity marker | maxpool(entity emb)+relu | micro | 0.86 | 0.86 | 0.86 | bert-large |
macro | 0.82 | 0.83 | 0.82 |
混合精度加速结果
在具体任务上,延续之前的setting,将train和dev合并共同作为新的train集,test集不变。在fp32 和fp16的两种setting下,比较相同batch_size下,一个epoch的用时或者每个迭代的用时。
比较方面 | fp32 | fp16 | 备注 |
---|---|---|---|
训练阶段 | 1.04it/s | 4.41it/s | 12.76it/s(独占显卡) |
推断阶段 | 4.14it/s | 8.63it/s | |
测试集指标 | 0.65/0.55 | 0.64/0.53 | 格式:micro/macro |
模型大小 | 421M | 212M |
Note that the project description data, including the texts, logos, images, and/or trademarks,
for each open source project belongs to its rightful owner.
If you wish to add or remove any projects, please contact us at [email protected].