Hi Everyone,
I wonder how we can implement these classical machine learning models into a live trading. In the courses, or in the most examples on the Internet, machine learning algorithms are developed based on fix datasets and hence, they are static models.
How we can dynamically use machine learning models that they incrementally optimise their algorithms as the new data arrives. In other words, no train-test split, just dynamic learning as the data arrives.
Maybe RL model can be converted to this as q network can be updated with the new data, but confused for other models including decision trees etc.
I would appreciate any help and thank you in advance 
Best Regards
Hi Suleyman,
That's an interesting question.
If you want to make it dynamic, you can retrain and test your model with newer data at a fixed frequency such as monthly, fortnightly or every time you are trading using the machine learning model. If while testing, the accuracy or any other metrics you prefer falls below a particular threshold level, you can stop trading using the ML model and when the accuracy comes back you can again start trading using the ML model.
You can refer to the below blog for more details and implementation using Blueshift
https://blog.quantinsti.com/machine-learning-strategy-using-blueshift-visual-programming/
I hope this helps.
Thanks
Hi Ishan,
Yes, this can work. Thank you so much for the reply.
Regards
Suleyman