[Trading Machine Project] Feedback

Dawn Filter

• 그동안 23시 50분 ~ 06시 00분사이의 데이터가 끊겨있던 것을 간과

Data Grouping

• [기존] N second Grouping -> Data Loss
• 2022-12-16 21:05:30 ~ 2023-02-25 23:49:00 (151838 rows)
• 2024-03-10 344252 rows, 2024-03-16 450793 rows
• 30s starting point problem
• Grouping 30s little bit dangerous
• [Reference] Use Whole Data (LOB model)

resample 1s

resample 30s

• but this also little bit dangerous
• averaging more than 100 data, a lot of data is lost

Data scaling

• [기존] Train / Validation / Test 를 Train의 평균과 표준 편차로 전처리를 진행
• [NEW] T(Lookback) 범위 단위로 Min-Max Scale

Time Series Model Repo

• https://github.com/khuangaf/CryptocurrencyPrediction/tree/master

• https://github.com/alphacsc/alphacsc

• convolutional dictionary learning

• 다변량 시계열 데이터에서 노이즈에 강건한 패턴 마이닝 할 때 유용