Machine learning in trading: theory, models, practice and algo-trading - page 923

 
Aleksey Vyazmikin:

I wanted to look at your data for model training now, to practice looking for tree parameters. But I couldn't find the files, they got lost in about twenty pages. Please attach them here again?

 
Dr. Trader:

I wanted to look at your data for model training now, to practice looking for tree parameters. But I couldn't find the files, they got lost in about twenty pages. Could you attach them here again, please?

Certainly it's possible, but I will have to do it in parts (the server is glitchy). Set Filter - determination of where no buy/sell entries, set MaloVhodov - trend entries for decent profit, set MnogoVhodov - all entries except loss ones.

I can't teach the tree to work out of sample. Of the predictors that safely affected the set MaloVhodov - target -1 I singled out the following:

arr_iDelta_H4

arr_iDelta_D1

arr_iDelta_MN1

arr_TimeH

arr_Den_Nedeli

arr_iDelta_Max_D1

arr_iDelta_Min_D1

arr_Regresor

arr_LastBarPeresekD_Down

arr_LastBarPeresekD_Up_M15

arr_LastBarPeresekD_Down_M15

arr_DonProc_M15

Files:
Filter.zip  3502 kb
 
continuation now set - MaloVhodov
Files:
MaloVhodov.zip  3471 kb
 
continuation now set - MnogoVhodov
Files:
MnogoVhodov.zip  3500 kb
 

1.

https://towardsdatascience.com/introduction-to-various-reinforcement-learning-algorithms-i-q-learning-sarsa-dqn-ddpg-72a5e0cb6287

2.

https://towardsdatascience.com/introduction-to-various-reinforcement-learning-algorithms-part-ii-trpo-ppo-87f2c5919bb9

It may be useful for Vladimir's article. For continuous tasks everything before DDPG is irrelevant, since there are tabular methods for a limited number of states/transitions

Introduction to Various Reinforcement Learning Algorithms. Part I (Q-Learning, SARSA, DQN, DDPG)
Introduction to Various Reinforcement Learning Algorithms. Part I (Q-Learning, SARSA, DQN, DDPG)
  • 2018.01.12
  • 黃功詳 Steeve Huang
  • towardsdatascience.com
Reinforcement Learning (RL) refers to a kind of Machine Learning method in which the agent receives a delayed reward in the next time step…
 
Maxim Dmitrievsky:

1.

https://towardsdatascience.com/introduction-to-various-reinforcement-learning-algorithms-i-q-learning-sarsa-dqn-ddpg-72a5e0cb6287

2.

https://towardsdatascience.com/introduction-to-various-reinforcement-learning-algorithms-part-ii-trpo-ppo-87f2c5919bb9

Perhaps Vladimir will be useful for the article. For continuous tasks everything before DDPG is irrelevant, since there are tabular methods for a limited number of states/transitions

Thanks. I'm bookmarked.I'll finish with ensembles (another article) and prepare on RL

Good luck

 
Maxim Dmitrievsky:

Proof:

After training we have a table graph like this: (from 04.01 OOS)


The 7th agent highlighted in yellow has the smallest errors. Let's drop all except him and see:

The result has improved.

Cool! Now (a day or two? sooner? as it goes...) I will finish one idea and get to your article!

 
Aleksey Vyazmikin:

Cool! Now (a day or two? sooner? as it goes ...) I will finish one idea and get to your article!

It would be nice, because the people I talked to did not offer any ideas, but just used what they were given.

and brainstorming is always useful

 
Dr. Trader:

Teach Max to flip the TS signals :)

And the pound is strictly in the red, while trading on the contrary it would be in the plus.

i understand - NS is a supersampling TS

But anyway, the forex broker knows in advance where TS will be buy and where it will be sell.

So everything goes according to plan and it is useless to reverse it.

 
Renat Akhtyamov:

However, all the same, the quotes are known in advance - where the TS will be buy and where it will be sell

Did you find out?