Two-Step Disentanglement for Financial Data

This is Pytorch implementation of "Two-Step Disentanglement for Financial Data". Using this method, we disentangled topic-wise and market-wise impact on Korean stock prices.

• "Clustering National Security Threats Using Two-Step Disentanglement Method On Stock Prices", Minh Choi, Sungwon Lyu, Sungzoon Cho, Korea Data Mining Society 2018 Fall Conference, Special Session Best Paper

Requirements

• Pytorch 0.4

Result

Loss

Accuracy of the first step(left) and loss of the second step(right).

Embedding of Zs

Embedded Zs of unstable periods(left) and of peaceful periods(right).

Analysis of Zs

The distribution of Zs of major provocations by North Korea.