技巧

在时序任务中，nn通过对特征的权重来拟合预测结果。如果时序数据中几乎没有类别变量的同质化数据，nn可能是第一选择。树模型本身是用split特征来预测，通过特征工程能更好的抓取复杂类别中蕴含的关系来做预测。

树模型

特征工程

深度学习

特征工程

阅读

• 有了Transformer框架后是不是RNN完全可以废弃了？ - sonta的回答 - 知乎

• Spatial-Temporal时间序列预测建模方法汇总

• https://github.com/Arturus/kaggle-web-traffic/blob/master/how_it_works.md

• https://github.com/sjvasquez/web-traffic-forecasting

• https://github.com/jfpuget/Kaggle

• https://github.com/jerrywn121/TianChi_AIEarth

• https://github.com/icodeworld/AI-competition/blob/main/tianchi-enso-prediction/RUN.py

• https://github.com/Wangjw6/Tianchi_Prediction

• https://github.com/Phimos/SIGSPATIAL-2021-GISCUP-2nd-Place-Solution

• https://github.com/tstanczyk95/WeatherGCNet

• https://github.com/vishalsinghroha/Short-Term-Wind-Speed-Prediction-based-on-Deep-Learning/blob/master/main.py

• https://github.com/alexalex222/Wind-Speed-Prediction

• https://github.com/sharmamayank1741/Wind-Speed-Prediction-using-Time-Series-Analysis/blob/master/LSTM_10_daily.ipynb

• https://github.com/Wizaron/deep-forecast-pytorch/blob/master/lib/model.py

• https://github.com/LenzDu/Kaggle-Competition-Favorita

• https://github.com/jingw2/demand_forecast

• https://github.com/zaburo-ch/kaggle-rrv-25th

reference

验证 https://www.kaggle.com/code/sergeifironov/validate-ridge

embed + NN (早期NLP NN比赛)

transformer 参考

• https://github.com/nadavbra/protein_bert/tree/master

• https://github.com/facebookresearch/esm

  • 使用transformers: https://www.kaggle.com/code/simonveitner/prepare-mean-embeddings

• https://github.com/Rostlab/goPredSim

• https://github.com/biomed-AI/SPROF-GO

  • https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/417561

  • https://www.kaggle.com/code/samusram/blastp-sprof-go

• https://github.com/SamusRam/ProFun

• https://github.com/agemagician/ProtTrans

• https://github.com/Shen-Lab/TALE

• https://github.com/kexinhuang12345/DeepPurpose

• 多目标参考kaggle-opm

不同的go term之间有图关系？ https://github.com/tanghaibao/goatools https://zhuanlan.zhihu.com/p/99789859 https://www.kaggle.com/competitions/foursquare-location-matching/discussion/336124

读物

https://zhuanlan.zhihu.com/p/571050734

关于测试集 https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/403200

关于parent term https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/403166

knn https://www.kaggle.com/code/geraseva/keops-knn

Stationarity and Memory in Financial Markets