Discussing the article: "Neural Networks in Trading: Dual Clustering of Multivariate Time Series (Final Part)"

New comment
 

Check out the new article: Neural Networks in Trading: Dual Clustering of Multivariate Time Series (Final Part).

We continue to implement approaches proposed vy the authors of the DUET framework, which offers an innovative approach to time series analysis, combining temporal and channel clustering to uncover hidden patterns in the analyzed data.

For model training, we use a dataset of historical data for the EURUSD currency pair on the M1 timeframe for the entire year 2024. During data collection, indicator parameters are kept at their default values.

Model training is conducted in two stages. First, the batch size is set to 1, so that a random state from the training dataset is selected at each iteration. This helps the model adapt to varying conditions. However, this is not sufficient for proper functioning of the risk management block. Therefore, in the second stage, the batch size is increased to 60, allowing sequences of 60 environment states and corresponding Actor actions to be taken into account. This makes the training process more stable and efficient.

The trained model is tested on historical data from January–February 2025. All settings are preserved, ensuring an objective evaluation of forecast quality. The testing results are presented below.


Author: Dmitriy Gizlyk

New comment