Intro
Using ML techniques is more challenging than it looks for some reasons including…
- Data Preprocessing:
- to understand data that comes from different sources with different formats
- to decide which parameters to use in the model
- Select Algorithm & Framework
- many problems appears to be similar, and there is no clear guideline for a particular problem
- Train & Tune Model
- need to feed data for many iteration, while tuning hyperparameters depending on the progress
- Integrate & Deploy
- need to setup an environment for deployment
- need to integrate the trained model into the environment
SageMaker:
- provide a single platform for all those steps (reduce complexity)
- we can interact with the system using Jupyter Notebook
- we can elastically deploy resources on demand