GNN upper edge distributor! Instead of trying to refine pills, you might as well give your GNN some tricks

author  | esang

Research direction  |  Figure neural network

2nd OGB-LSC Already open , You don't know this one yet GNN Do you use sharp tools ？

gtrick It's easy to use , oriented GNN Of trick hold-all ：

https://github.com/sangyx/gtrick

idea

The of this project idea From the right to OGB Observations on the list , You will find that although the model changes , But everyone's on the trick But it's limited . In that case , Why not just pack these trick Make a library convenient for everyone to use ？

Besides ,GNN The task scenarios are relatively concentrated , The input is also relatively fixed , It is very suitable for implementing some plug-in trick, There is no need to change the model . Based on such an idea ,gtrick emerge as the times require .

advantage

gtrick There are three commendable advantages ：

1. Simple and easy to use . As shown in the figure below , Simply add a few lines of code to add a trick Introduce your GNN In the model , Make as few changes as possible to the existing code .

▲  Original model

▲  introduce trick: Random Feature

2.  Every Trick Will be verified . We will select a dataset pair trick To verify , Only those that can really improve the performance of the model on the test data set trick Will be included in gtrick in .

▲  Some collections trick stay ogbn-arxiv Test results on

3. Support at the same time DGL And PyG. We treat each trick At the same time provide DGL And PyG Two implementations and corresponding examples . No matter what kind of in-depth learning library you like , You can try it .

▲ gtrick The implemented 10 individual trick And provide corresponding examples

Besides , It is worth noting that , Although some trick Already in PyG or DGL To realize , such trick We will still include gtrick And provides another implementation of graph depth learning library , because gtrick The first goal of is to let everyone know that there are some methods that can really and effectively improve the performance of the model in various models .

install

at present gtrick Already included 10 For different tasks trick, And through pip Direct installation ：

pip install gtrick

at present gtrick Still under development , And all kinds of implementations are rough . You can also direct the corresponding trick Add your code to your project .

Last , Welcome to use , Welcome to make suggestions , Warmly welcome to order star：

https://github.com/sangyx/gtrick

Read more

# cast draft   through Avenue #

  Let your words be seen by more people  

How to make more high-quality content reach the reader group in a shorter path , How about reducing the cost of finding quality content for readers ？ The answer is ： People you don't know .

There are always people you don't know , Know what you want to know .PaperWeekly Maybe it could be a bridge , Push different backgrounds 、 Scholars and academic inspiration in different directions collide with each other , There are more possibilities . 

PaperWeekly Encourage university laboratories or individuals to , Share all kinds of quality content on our platform , It can be Interpretation of the latest paper , It can also be Analysis of academic hot spots 、 Scientific research experience or Competition experience explanation etc. . We have only one purpose , Let knowledge really flow .

  The basic requirements of the manuscript ：

• The article is really personal Original works , Not published in public channels , For example, articles published or to be published on other platforms , Please clearly mark  

• It is suggested that  markdown  Format writing , The pictures are sent as attachments , The picture should be clear , No copyright issues

• PaperWeekly Respect the right of authorship , And will be adopted for each original first manuscript , Provide Competitive remuneration in the industry , Specifically, according to the amount of reading and the quality of the article, the ladder system is used for settlement

  Contribution channel ：

• Send email ：[email protected] 

• Please note your immediate contact information （ WeChat ）, So that we can contact the author as soon as we choose the manuscript

• You can also directly add Xiaobian wechat （pwbot02） Quick contribution , remarks ： full name - contribute

△ Long press add PaperWeekly Small make up

Now? , stay 「 You know 」 We can also be found

Go to Zhihu home page and search 「PaperWeekly」

Click on 「 Focus on 」 Subscribe to our column

·

·

·