API: developer interface (user interface의 반대 위치.)
Developer input -> Code -> Output for developer
good API?
- Stable
- Integrates with existing tools
- Intuitive
- Flexible/extendable
Data -> ETL -> Model(Train -> Validate -> Predict)
ETL: extract transform and load. All of pre-processing
- taking in some data and then returning different data which is hopefully better
- Data -> Categorical expansion -> Null imputation -> Feature scaling -> Better data
- Transformers: Categorical expansion -> Null imputation -> Feature scaling
Train: Data -> Transformer -> Model(Estimator.fit(x,y), estimator.predict(x))
Validation: Data -> Transformer -> Model(Estimator)
- Meta-estimator: Transformer -> Model(Estimator)
Pipeline: something that takes in a series of steps
- Transformer(s) -> Stacking(Base 1, 2, 3... -> Model)
Model stacking: to find a model that works well, smash models all together and get one mega better model
microCT_applications