Machine learning in trading: theory, models, practice and algo-trading - page 1299
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
Yes, I did something similar - the question is again about predictors and selection criteria (target). Now (many months later) I will finish all ideas with predictors and return to this topic. And the result is in general, previously posted how similar models work, but we need a variety of samples with different scatter, preferably from different models.
And what does this AutoML uses as predictors and targets?
well, the target must be known, and predictors are transformed on the machine, and the models are enumerated
I'll write when (and if) will learn more
I'm waiting when googles will update TensorFlow to 2.0, hopefully with support for python 3.7. I like everything from google and this package is the one for all occasions, it has everything.
well, the targets should be known, and the predictors are transformed on the machine, more models are taken over
I will write when (and if) I learn more
I am waiting when google will update TensorFlow to 2.0, hopefully with support for python 3.7. I like everything from google, and this package is the one for all occasions, it has everything.
Write when you figure it out, very interested in the predictors if they are publicly stipulated there. The target is also difficult in our case because there is an additional estimation of the model on money, it is good for fixed TP and SL, in other cases even a good model from the point of view of predictive ability may fail. Anyway I take into account the curve of classification balance in model estimation and check it for drawdowns and other criteria like for usual balance because I expect uniformity of classification accuracy in all sample.
I don't have python yet, I can do a lot of things without it, I have a lot of ideas that need to be realized.
It is very interesting to see the predictors, if they are publicly stipulated there. It is good with fixed TP and SL, in other cases even a good model from the point of view of predictive ability can fail. Anyway I take into account the curve of classification balance in model estimation and check it for drawdowns and other criteria like for usual balance because I expect uniformity of classification accuracy in all sample.
I have not yet put python - I can do a lot of things without it, I have a lot of ideas that need to be implemented.
I mean, predictors are also your own, but transformed by AutoML itself and so are i.e. automatic selection.
I mean the predictors are also yours, but they are transformed by AutoML itself and the selection is correspondingly automatic.
There's nothing interesting in essence, it's just a wrapper providing additional functions :)
Then essentially nothing interesting, just a wrapper that gives extra features :)
It was about getting rid of the routine
As opposed to the fact that some articles suggest doing all datamining routine manually which is absurd in non-stationary markets
I've already written many times my opinion that the statistical approach does not work in non-stationary markets (i.e. the classical one described in books on statistics and MO)It was about getting rid of the routine
As opposed to the fact that some articles suggest doing the whole datamining routine manually, which is absurd in non-stationary markets
I've already written many times my opinion that statistical approach does not work in non-stationary markets (i.e. classical, described in books on statistics and MO)Well, you still have to come up with everything yourself - targets and predictors. I thought there was some research about finding signs of an over-trained model by its structure or something. It is important for me to learn exactly from the data of training and test samples to find a model that will work on a sample independent of training (or on the contrary - which exactly will not work), but here also raises the question "what does it mean to work?" and it is not unambiguous. And automation of model analysis is not hard, at least in catbust all you need is unloaded into different files, and then just parse them, either with MT, as I do, or with your own software.
Well, you still have to come up with everything yourself - targets and predictors. I thought there was some research about finding signs of an over-trained model by its structure or something. It is important for me to learn exactly from the data of training and test samples to find a model that will work on a sample independent of training (or on the contrary - which exactly will not work), but here also raises the question "what does it mean to work?" and it is not unambiguous. And automation of model analysis is not difficult, in any case in catbust all you need is unloaded into different files, and then just parse them, either with MT, as I do, or with your own software.
it seems that way to you now, because you haven't tasted sweeter carrots yet.
You will get there in time, if you study instead of fantasizing...
Because your fantasy, compared to the fantasy of Google or dipmind teams in the field of AI, is a drop in the ocean. That's why you have to take ready-made.
If they write that the model should be used so-and-so, then so-and-so is the net. There's nothing to fantasize about, because it's important to understand what has already been invented.It seems that way to you now, because you have not yet tasted sweeter carrots
you will get there in time, if you study instead of fantasizing.
Because your fantasies like those of Google or Dipmind, in the field of AI, are a drop in the ocean. That's why you have to take the ready-made
I do not understand the depth of thought. You wrote that the target and predictors need to come up with your own for that software, and if so, then I wrote that then there's not much point in it, because you can do everything yourself and will know how it works, what to improve and fix.
Regarding the second part of the statement, I agree, but I prefer to take what I understand how it works.
I did not understand the depth of thought. You wrote that both the target and predictors must come up with your own for that software, and if so, then I wrote that then there is not much point in it, because you can do everything yourself and will know how it works, what to improve and fix.
Regarding the second part of the statement, I agree, but I prefer to take what I understand how it works.
AutoML was cited as an example of automating the entire process (or almost everything), that's the depth of thought. The chain of logic is straight from the initial post. You are leading yourself into a dead end.
The example was given to explain what is meant by generalization, not manual extraction of knowledge, as you have. It even says that the machine does better than the man in almost all phases.
I doubt you understand how catbust works.
autoML was given as an example of automating the whole process (or almost the whole process), that is the depth of thought. The logical chain is straight from the initial message. You are leading yourself into a dead end.
The example was given to explain what is meant by generalization, not manual extraction of knowledge, as you have. It even says that the machine does better than the man in almost all phases.
I doubt you understand how catbust works.
Okay, try it, tell me if you want to. And if possible, compare my selection and ML selection, if at all possible without too much work on my part.
Yes, I don't thoroughly understand how a catbust works, but I already have the knowledge and experience of the operation, and all that takes time, which leads to a comprehensive understanding. To take something from scratch and try to apply in the work, when there is no sufficient information, well, for me it is not comfortable. Even with catbust I have to search and understand everything, translate, so it's good that I have people who understand it better, who I can ask technical aspects of code.