Discussing the article: "Neural Networks in Trading: Transformer for the Point Cloud (Pointformer)"

New comment
 

Check out the new article: Neural Networks in Trading: Transformer for the Point Cloud (Pointformer).

In this article, we will talk about algorithms for using attention methods in solving problems of detecting objects in a point cloud. Object detection in point clouds is important for many real-world applications.

After several iterations of model training and dataset updates, we succeeded in obtaining a policy that is capable of generating profit on both the training and test datasets.

We evaluated the performance of the trained model using the MetaTrader 5 Strategy Tester, running tests on historical data from January 2024, while keeping all other parameters unchanged. The test results are presented below. 

During the test period, the trained model executed a total of 31 trading operations, half of which were closed in profit. Notably, a nearly 50% higher value in maximum and average profitable trades compared to their losing counterparts led to a profit factor of 1.53. Despite the upward trend observed in the equity curve, the limited number of trades prevents us from drawing any definitive conclusions about the model’s effectiveness over a longer time horizon.



Author: Dmitriy Gizlyk

 
Dmitriy Gizlyk :
on historical data from January 2024.

Why only January, isn't it already September? Or is it implied that one has to retrain every month?

 
Aleksey Vyazmikin #:

Why only January, is it already September? Or is it implied that one has to retrain every month?

You can't train a model on 1 year of data and expect stable performance over the same or longer time frame. To get stable model performance for 6-12 months, you need a much longer history to train. Consequently, it will take more time and cost to train the model.

New comment