Machine learning in trading: theory, models, practice and algo-trading - page 1806
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Hi Folks.
I am using Deep Reinforcement Learning (using Python) and learning is already happening (as shown):
Today I am using just moving averages as a market observation for learning.
This model performs several actions (buy, sell, and wait). Thus, after training, the model converges in many actions to "wait" with only the best actions.
However, the training in this model is very slow because it uses all the ticks.
What do you suggest as market observation data to improve accuracy and reduce losses?
Sorry for my translation errors.
What do you suggest as market observation data to improve accuracy and reduce losses?
1) Create a market model, and train the agent in it, it will reduce the dimensionality and accelerate training, even here so did
2) Selection of features, it is as they say a creative task, if there are a lot of them then it will also help to reduce dimensionality, there are a lot of them from clustering to pca,umap and so on.
If we take support and resistance levels and decide to buy or sell only if the price is at that level, then we can decrease the training sample by orders of magnitude.
You can combine all the points together .
Hi Folks.
I am using Deep Reinforcement Learning (using Python) and learning is already happening (as shown):
Today I am using just moving averages as a market observation for learning.
This model performs several actions (buy, sell, and wait). Thus, after training, the model converges in many actions to "wait" with only the best actions.
However, training is very slow in this model because it uses all the ticks.
What do you suggest as market observation data to improve accuracy and reduce losses?
Pardon my translation errors.
Thin out ticks to acceptable error.
Identify areas where the error exceeds the allowable error when removing all ticks.
Train the model to find such areas, if it works out of course))))
Well, these are state space models that work every once in a while, too
And you can bind the target to a change in the trend. The task is not for profit, but for recognition and prediction. It is clear that completely different tasks, NS is not able to recognize and predict at the same time)))) Target profit, balance recognize pivot points indirectly, without direct recognition.
If we take the zigzag reversals daily, 4h, hourly (or whatever), and look for example on each (all) TF 120 bars and 120 ticks before and after the reversal point. The daily and 4h watch only up to the next and previous reversal or do not know how to do it better. Something needs to be written manually then of course, to prepare the data.
Or is there something similar.
And you can tie the target to a change in the trend. The task is not for profit, but for recognition and prediction. It is clear that completely different tasks, NS is not able to recognize and predict at the same time)))) Target profit, balance recognize pivot points indirectly, without direct recognition.
If we take the zigzag reversals daily, 4h, hourly (or any other), and look for example at each (all) TF 120 bars and 120 ticks before and after the reversal point. The daily and 4h watch only up to the next and previous reversal or do not know how to do it better. Something needs to be written manually then of course, to prepare the data.
Or is there something similar.
Hi Folks.
I am using Deep Reinforcement Learning (using Python) and learning is already happening (as shown):
Today I am using just moving averages as a market observation for learning.
This model performs several actions (buy, sell, and wait). Thus, after training, the model converges in many actions to "wait" with only the best actions.
However, training is very slow in this model because it uses all the ticks.
What do you suggest as market observation data to improve accuracy and reduce losses?
Sorry for my translation mistakes.
I think deep reinforcement just used for a multivariate tasks, not for 1-5 dimensional, like financial markets. So you can try more simple and faster methods like REINFORCE with neural network or linear approximator.
This is very easily checked by the presence of regularity between zigzag breaks, i.e. entropy reduction. If there will be a difference from sb, then you can look. But I have not found such things. There are special packages, used in medicine to analyze dna and derive drug formulas. For example permutation entropy, or more complicated cases with approximators, optimizers like q-learning
Nah, fracture regularity doesn't work)))) Something before a fracture needs to be caught without considering regularity) In general there is a chance to catch those fractures that have the same differences from the SB before the fracture. You can't do that with regular packages. And I don't think there is anything significant for a break in the middle of the trend, except too long trend duration))))
Nah, the regularity of fractures doesn't work)))) Something before the fracture must be caught, not considering the regularity) In general, the chance is to catch those fractures in which there are the same differences from the SB before the fracture. You can't do that with regular packages. And I don't think there is anything meaningful for a break in the middle of the trend, other than the trend being too long))))
Well I see, by analogy with TF correlation. I can try
I think deep reinforcement just used for a multivariate tasks, not for 1-5 dimensional, like financial markets. So you can try more simple and faster methods like REINFORCE with neural network or linear approximator.
If you look at different TFs you will get more dimensions))))
I see, by analogy with TF correlation. We can try.
Yes. Only TFs would be all to see, which are needed at once can not be determined. Maybe a neighboring one, or maybe one or two TFs will be better, or maybe all of them are needed))). It is hard to think it through.)