标准流程：

1. 理解问题：理解问题的核心，相关领域的问题、经验、限制、惯例、内外影响等等。
2. Collect input features
3. Preprocess：Fillna(fill 0,mean,or by model(eg rf)), Outlier
4. Feature engineering：
1. Normalize: min-max,z-score,pca,zca
2. Transform: square,log,exp,sin,cos,rotate
3. Embedding: one-hot, category
4. Binning: eg. age 0-14:1 , 14-20:2
5. Cross feature: eg. X1*X2
6. De-periodic：eg. fft
7. TD: y[n] = x[n] -x[n-t]
5. Sampling: Uniform, Stratified, Pool, Undersampling, Oversampling,MCMC, Gibbs, SMOTE
6. Build Model : DL or ML
7. Train: Hyper params(grid search), cross validation
8. Validate: Get metrics