Kaggle Completion Note

[NeroCube]

ENV

• Align the package version (pandas, numpy, sklearn)

  Slack

• All the discus on the platform
• Explain what you will do
• Set a check point (1w)

  Baselines

• Put baseline version on Github
• Use the same seed

  Tips

• bagging > diff seeds
• pseudo labeling
• metric
• weights
• different models
• embedding
• create features
• change y to square
• add bias
• change y to bins
• predict diff
• if improve method work, change seed that should still work
• feature different windows diff, rolling windows mean, sum, max, log return, rv
• generate large seed and blend them
• save train, test to pickle to reduce preprocess time