ML lifecycle architecture

---

Global Trade Data provides global import/export data and analytics.  They need predict future price of a specific product, such as shoes.  Many factors can affect the price of shoes, such as related raw materials, electricity cost, tariff, exchange rate, etc.

Dimension reduction; Non-linear regression (SVM, MLP)

The innate flaw of the first solution was the temporal dependencies cannot be captured. LSTM/RNN was used to solve this problem.

The revolutionary attention/transformer architecture replaced LSTM and improved the accuracy of prediction.