All Projects → plantsgo → Ijcai 2018

plantsgo / Ijcai 2018

ijcai-2018 top1 solution

Projects that are alternatives of or similar to Ijcai 2018

Production Data Science
Production Data Science: a workflow for collaborative data science aimed at production
Stars: ✭ 388 (-1.77%)
Mutual labels:  jupyter-notebook
Workshopscipy
A workshop for scientific computing in Python. ( December 2017 )
Stars: ✭ 391 (-1.01%)
Mutual labels:  jupyter-notebook
Examples
Example deep learning projects that use wandb's features.
Stars: ✭ 391 (-1.01%)
Mutual labels:  jupyter-notebook
Mlcourse
Machine learning course materials.
Stars: ✭ 388 (-1.77%)
Mutual labels:  jupyter-notebook
Private Ai
Repo for Udacity's Secure & Private AI course
Stars: ✭ 391 (-1.01%)
Mutual labels:  jupyter-notebook
Pattern classification
A collection of tutorials and examples for solving and understanding machine learning and pattern classification tasks
Stars: ✭ 3,880 (+882.28%)
Mutual labels:  jupyter-notebook
Bayesian changepoint detection
Methods to get the probability of a changepoint in a time series.
Stars: ✭ 387 (-2.03%)
Mutual labels:  jupyter-notebook
User Machine Learning Tutorial
useR! 2016 Tutorial: Machine Learning Algorithmic Deep Dive http://user2016.org/tutorials/10.html
Stars: ✭ 393 (-0.51%)
Mutual labels:  jupyter-notebook
Open source demos
A collection of demos showcasing automated feature engineering and machine learning in diverse use cases
Stars: ✭ 391 (-1.01%)
Mutual labels:  jupyter-notebook
Deepnetsforeo
Deep networks for Earth Observation
Stars: ✭ 393 (-0.51%)
Mutual labels:  jupyter-notebook
Graph transformer networks
Graph Transformer Networks (Authors' PyTorch implementation for the NeurIPS 19 paper)
Stars: ✭ 387 (-2.03%)
Mutual labels:  jupyter-notebook
Keras Facenet
Facenet implementation by Keras2
Stars: ✭ 390 (-1.27%)
Mutual labels:  jupyter-notebook
Nmtpytorch
Sequence-to-Sequence Framework in PyTorch
Stars: ✭ 392 (-0.76%)
Mutual labels:  jupyter-notebook
Learning Deep Learning
Paper reading notes on Deep Learning and Machine Learning
Stars: ✭ 388 (-1.77%)
Mutual labels:  jupyter-notebook
Megvii Pku Dl Course
Homepage for the joint course of Megvii Inc. and Peking University on Deep Learning.
Stars: ✭ 393 (-0.51%)
Mutual labels:  jupyter-notebook
Latex ocr pro
🎨 数学公式识别增强版:中英文手写印刷公式、支持初级符号推导(数据结构基于 LaTeX 抽象语法树)
Stars: ✭ 385 (-2.53%)
Mutual labels:  jupyter-notebook
Tensorflow Docs
TensorFlow 最新官方文档中文版
Stars: ✭ 3,782 (+857.47%)
Mutual labels:  jupyter-notebook
Motion Cosegmentation
Reference code for "Motion-supervised Co-Part Segmentation" paper
Stars: ✭ 393 (-0.51%)
Mutual labels:  jupyter-notebook
Baiduyun deeplearning competition
百度云魅族深度学习应用大赛
Stars: ✭ 393 (-0.51%)
Mutual labels:  jupyter-notebook
Zhihu Text Classification
[2017知乎看山杯 多标签 文本分类] ye组(第六名) 解题方案
Stars: ✭ 392 (-0.76%)
Mutual labels:  jupyter-notebook

ijcai-2018

ijcai-2018 top1 solution 一.Graft Learning 首先科普一下什么是嫁接。

在生产实践过程中,嫁接对经济价值的提高,也有着非常多的实例:如普通的水杉,价值一元;而通过嫁接手段,培育成金叶水杉后,经济价值提高20多倍;再如普通的大叶女贞树,价值几角钱;而通过嫁接的手段,培育成彩叶桢树后,其经济价值更高达近百倍。由此可见,嫁接对品种的改良,经济价值的提高都有首非常重要的意义。

影响嫁接成活的主要因素是接穗和砧木的亲和力,其次是嫁接的技术和嫁接后的管理。所谓亲合力,就是接穗和砧木在内部组织结构上、生理和遗传上,彼此相同或相近,从而能互相结合在一起的能力。亲和力高,嫁接成活率高。反之,则成活率低。一般来说,植物亲缘关系越近,则亲和力越强。例如苹果接于沙果;梨接于杜梨、秋子梨;柿接于黑枣;核桃接于核桃楸等亲和力都很好。

大家应该都尝试过只用只用前七天的数据来预测,最后一天来做线下验证,这种效果不太好,但是数据量大,就如同这里的砧木获取容易,但是实际价值不大。 大家应该也尝试过只用最后半天来做预测,达到的效果也已经很不错了,并且比只用前七天的训练好,但是数据量比较少。

那么我们参考嫁接的思想,让砧木来给我们的接穗输送营养。在这里前七天的数据就类似于砧木,后半天的数据类似于接穗。前七天预测出来的结果就是营养。这样我们预测出来的结果也更加接近于最后一天。

那么既然这样我为什么没有使用初赛的数据来做“嫁接“呢? 亲缘关系越近的嫁接成功率越高,在这个问题上也是item相同品别的做这种“嫁接“操作效果才会好,而我们的初赛数据和复赛数据的item并不是一个品类的,所以价值不大。

二.Sample Embedding sample_emb_x=[x1,x2,x3,x4,...,xn] xn为第n个property在不在predict_category_property中 sample_emb_y=[y1,y2,y3,y4,...,yn] yn为第n个property在不在item_property_list中

一个user有很多个不同的item交互样本,一个item也有很多不同的user交互样本 user_emb_x=mean([sample_emb_x_1,sample_emb_x_2,...,sample_emb_x_k]) sample_emb_x_k为该user的第k条样本的sample_emb user_emb_y=mean([sample_emb_y_1,sample_emb_y_2,...,sample_emb_y_k]) sample_emb_y_k为该user的第k条样本的sample_emb

通过这种对所有样本的sample_emb做mean操作来对user做embedding item_emb_x=mean([user_emb_x_1,user_emb_x_2,...,user_emb_x_k]) user_emb_x_k为该item的第k条样本的user_emb item_emb_y=mean([user_emb_y_1,user_emb_y_2,...,user_emb_y_k]) user_emb_y_k为该item的第k条样本的user_emb

通过这种对所有样本的use_emb做mean操作来对item做embedding 至此通过predict_category_property,item_property_list这两条信息对sample,user,item做了embedding。 得到了6n个特征,n的大小视情况而定,这里我取了出现次数top100的property来做我的embedding,所以总共6100个特征。

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].