Discussion of article "CatBoost machine learning algorithm from Yandex with no Python or R knowledge required" - page 6
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
Maybe try the settings from the article?
It seems that I had to set the capital from 10k USD to 200k USD so now I have at least better results: Score of 24 for 15k trades with 0.89 PF
In my code set the lot size equal to one. You consider the code as a template for experimenting with CatBoost.
In my code set the lot size equal to one. You consider the code as a template for experimenting with CatBoost.
thanks, it looks better now but from 40 seeds there was still no higher than 0.5. I try to do more seeds. Is it different if it goes 1 to 100 by 1 or 1 to 10000 with a step of 100?
Do the quantization, and only then apply the seed. Each seed is different.
Do the quantization, and only then apply the seed. Each seed is different.
Thanks, I am not sure I fully understand you. Quantization and seeding is prepared in one step when setting it in the CB_bat script, right?
Anyway during night some 200-300 seeds were generated with better results, also mqh files were generated. When I backtested during the training period the equity curve nicely picking up but when I test it during the test and exam period trades are rarely taken. Ma period was 96 so now I started again from the beginning. Switched to DJI30 (for a change) used period 8 and M2 and optimized the price and MA type only. This way much more than 15k trades are generated (I even reduced the length of the period as the XXXCB_Save_pred.csv file is around 1.3Gb and 1 training cycle is 13 minutes. I set the seed parameter fro 1 to 10000 with a step of 100 which gives around 100 model. I hope there will be some result after this.
Thanks, I am not sure I fully understand you. Quantization and seeding is prepared in one step when setting it in the CB_bat script, right?
Anyway during night some 200-300 seeds were generated with better results, also mqh files were generated. When I backtested during the training period the equity curve nicely picking up but when I test it during the test and exam period trades are rarely taken. Ma period was 96 so now I started again from the beginning. Switched to DJI30 (for a change) used period 8 and M2 and optimized the price and MA type only. This way much more than 15k trades are generated (I even reduced the length of the period as the XXXCB_Save_pred.csv file is around 1.3Gb and 1 training cycle is 13 minutes. I set the seed parameter fro 1 to 10000 with a step of 100 which gives around 100 model. I hope there will be some result after this.
I recommend that you first find the best way to quantize by going through the different options, and then go through the seed. Ideally, you should search for your own quantization settings for each predictor, and then combine the results. Perhaps I will write about this in the next article.
Attached is the EA code and the script that organises the whole infrastructure - it was important to me that what I was describing could be reproduced - so test and report bugs, suggest improvements - I'm all for general input on this.
Thanks for the useful article!
Is it possible to leverage the GPU to train the model?
In a "head-on" approach, adding the option "--task-type GPU" to _01_Train_All results in an error: Error: change of option sampling_frequency is unimplemented for task type GPU and was not default in previous run
Thanks for the helpful article!
Is it possible to use GPU to train the model?
When adding the option "--task-type GPU" to _01_Train_All, the following error occurs: Error: change of option sampling_frequency is unimplemented for task type GPU and was not default in previous run
Unfortunately, I don't have an Nvidia card, so it's hard to figure out the cause. Start by removing keys as much as possible, not all features implemented on CPU are supported on GPU.
Unfortunately I don't have a card from Nvidia, so it's hard to figure out the cause. Start by removing keys as much as possible, not all features implemented on CPU are supported on GPU.
Alexey, good afternoon.
Thank you very much for the article!
Could you please tell me if it is possible to teach the AI attached to the article to calculate and draw a zigzag ?
Thank you very much for your reply.
Alexei, good afternoon.
Thank you very much for the article!
Could you please tell me if it is possible to teach the AI attached to the article to calculate and draw a zigzag?
Thank you very much for your reply.
Hello.
If we are talking about replacing an indicator, it is possible to train it, but if we are talking about predicting extrema, it is much more difficult, but the CatBoost algorithm allows you to do it, if there are corresponding predictors.