Machine learning in trading: theory, models, practice and algo-trading - page 405
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
Well, the problem from the 1st post I think everyone is interested in checking their MO system.
Interesting generator you have here. Is it self written?
Did you divide the plot into 2 or 3. What do you get with unfamiliar data?
This is a continuation of the EA generator that is described in the article, towards the forest).
But practical experiments on this data, I think, do not make much sense, because the example seems to be purely academic.
This is a continuation of EA generator, which is described in the article, towards the forest).
But experiments on this data, I think, don't make much sense, as the example seems to be purely academic.
Made also 1 derovo, which completely remembered the input file, but all bad on new data:
Average error on the training (60.0%) plot =0.000 (0.0%) nTrees=1 codResp=1
Average error on validation (20.0%) plot =0.706 (70.6%) nTrees=1 codResp=1
Average error in the test (20.0%) section =0.701 (70.1%) nTrees=1 codResp=1
So it makes sense
I suggest arranging a contest for the models of the MO, the models may be anything, even if they are cobbled together, it doesn't matter - the main thing is to have an element of learning / selecting weights / optimizing. You can have some fun with the demos. That would not be useless arguments what works and what does not:)
Truth is born in arguments)The problem is that many people here are prone to make unfounded statements - that all this is nonsense and will not work, but it should not be so and so on... without confirming their words with almost anything. I.e. the person either just rubs his tongue, not understanding about what, or really understands and he has results...
The problem is that many people here are prone to make unfounded statements, saying that all this is nonsense and will not work, but it should not work this way and so on... without confirming their words with almost anything. I.e. the person either just scratches his tongue without understanding what he is talking about, or really understands and has results...
I propose to arrange a contest for MO models, the models can be anything, even self-made, it does not matter at all... the main thing is that there would be an element of training / selection of weights / optimization. You can have some fun with the demos. So as not to engage in useless arguments about what works and what does not.) This way it will be clear at once who really understands how MI may be applied in markets and who is just babbling :)
I prefer practical results and I follow them, for example I put working indicators and Expert Advisors in market and if their code is created by generator without any additional programming, then it's free.
No, I just mean that Michael is attacked, for example, that he was using some Reshetov crap... and they do not offer a replacement and do not even fully understand it :)
I compared models - it produces a smaller percentage of errors on the test sample, what more do you need any proof :) At least it's better than woods and uncomplicated MLP, and how to apply it is already the 3rd question.
Then he'll make an awesome monitoring and he'll tear all theorists to pieces :D
No, I just mean that Michael is attacked, for example, that he was using some Reshetov crap... and did not offer a replacement and do not even fully understand it :)
I compared models - it produces a smaller percentage of errors on the test sample, what more do you need any proof :) At least it's better than woods and uncomplicated MLP, and how to apply it is already the 3rd question.
Then he'll write an awesome monitoring tool and rip all the theorists to shreds :D
Maybe there is something interesting in this system?
If you figured out this new model - tell me how he chooses the right ones out of 100 inputs, to feed two RNNs (there are 8 inputs each, right?)? A complete search of 8 out of 100? With the calculation of each and choose the best one? Or genetics or some other type of selection?
I don't know how exactly the selection happens, but there is one rule, as soon as the model is retrained, there is a complication of the model. With any training, the number of inputs is always different. It all depends on the sampling split for training and test by randomization....
This is the second day the model is counting..... That's what it means when you remove trash from a set.... :-)
If you understand this new model - tell me how it chooses the right inputs out of 100, to feed two RNN (there are 8 inputs, right?)? A complete search of 8 out of 100? With the calculation of each and choose the best one? Or genetics or other type of selection?
I haven't figured out how it sifts out uninformative predictors, I'm in the process of rewriting it on mkul
You will turn into a dragon until you rewrite and understand something, that's why I wrote that 2 weeks at the earliest)