search

yaroslavmavliutov / Time_series_Prediction-Anomaly-detection_V-AE-LSTM

9 stars 2 forks source link

ask for something about data #1

Open suesueyin opened 3 years ago

suesueyin commented 3 years ago

Can you push the data ? and what is the data's format?

yaroslavmavliutov commented 3 years ago

I used Index S&P500 and FOREX EUR/USD. You can easily find these data (.csv files) on the internet

suesueyin commented 3 years ago

Ok,thank you very much. I find the former one dataset. But I am engaging unlabeled and multi-dimensional data in time series anomaly detection model, and I really want to confirm what if your method could be used to solve my model. Sorry to bother you, I am looking forward to your reply

yaroslavmavliutov commented 3 years ago

My model is implemented for unlabeled data. I used models just for one feature, you can try to redefine input shape. Try, good luck

suesueyin commented 3 years ago

hey guy, I have tried your model in my multi-dimension data and it works so amazing! And I found actually your method is so called VAE network, then I want to ask about the GAN and VAE, which method is more efficient to solve time series anomaly detection problem? I am looking forward to your reply

yaroslavmavliutov commented 3 years ago

Hello,

Yes, i implemented here two types of VAE (standard and variational).

GAN is so powerful network, used to generate different content. I don't know how it works with financial data, you can do a research to find more details. For example (https://arxiv.org/pdf/1809.04758.pdf)

GAN takes more time to implement, but it can provide best result.