Machine learning in trading: theory, models, practice and algo-trading - page 2390
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
I completely forgot, that you have classes responsible for trade direction, while I have classes responsible for permission/prohibition to trade - that's why you can't feel the usefulness of the chart :)))
So the search for patterns is solved by enumerating the diversity of relations between predictors, just the increments are unstable and need to expand the range, at least add them ATR(3) daily.
But not in the same barbaric way
wait a week until it counts, then you have to strain your eyes to look at pictures
It's much easier to agree that there are no regularities)
I completely forgot, that you have classes responsible for trade direction, while I have classes responsible for permission/prohibition to trade - that's why you can't feel the usefulness of the chart :)))
If I wanted to handle it I would have to look for patterns by enumerating the various correlations of predictors, just the increments are unstable and it would be necessary to widen the range, at least add a daily ATR(3) to it.
And if there is a training on M1? For 1440 bars will be the same value of this indicator. Or you are training on D1?
but not in the same barbaric ways.
Wait a week until it counts, then break my eyes to look at picturesThere is a preselection by a set of criteria, and the graphs are more needed to understand the quality of the model on a particular network. Of course, the graphs will vary depending on the target.
For example, if the model generally shows the profit but there are strong profit failures in the middle of the probability, for example by 0.6, I won't take such a model, and if these failures are at the ends of distribution, I can simply limit the response (interpretation of the unit), say, to 0.65.
Although it can be seen that the model itself is not very good (no pronounced two humps), compared to the last one.
ATR on D1 with a period of 3?
And if on M1 training? For 1440 bars will be the same value of this indicator. Or you are training on D1?
Yes, it will be one and the same - it's volatility definition and the model should define for example 2-3 volatility periods, on which values in pips should be interpreted differently, because for some ranges it is the beginning of the trend, and for others it is already the end. Well, I also write such values simply in the ATR, so chunks with different volatility become comparable.
There is a preselection by a set of criteria, and the graphs are rather needed to understand the quality of the model on a particular network. The graphs, of course, will vary depending on the target.
For example, if the model generally shows the profit but there are strong profit failures in the middle of the probability, for example by 0.6, I won't take such a model, and if these failures are at the ends of distribution, I can simply limit the response (interpretation of the unit), say, to 0.65, as here.
Although you can see that the model itself is not very good (no pronounced two humps), compared to the last one.
This is all bullshit, we need new breakthrough ideas
I wouldn't lift a finger without themI tried yesterday to make a generative-adversarial algorithm based on your idea from the video. There is a generator agent that grinds deals, and there is a discriminator agent that evaluates the correctness of deals and removes the negative ones. Dataset was selected by a sliding window with discrete step. Unfortunately, I have not yet managed to get a stable learning process, because at 5-7 iterations the discriminator deleted the whole dataset))). I tried resampling before training and both agents according to your idea, but not very much. I'm going to try tonight to reverse or randomize the trades instead of deleting them. I understand that removing invalid tags is more efficient than modifying or randomizing them, but I'd like to run a non-reversible learning process.
I tried yesterday to make a generative-adversarial algorithm based on your idea from the video. There is a generator agent that grinds deals, and there is a discriminator agent that evaluates the correctness of deals and removes negative ones. Dataset was selected by a sliding window with discrete step. Unfortunately, I have not yet been able to get a stable learning process, because at 5-7 iterations the discriminator deleted the entire dataset))). I tried resampling before training and both agents according to your idea, but not very much. I'm going to try tonight to reverse or randomize the trades instead of deleting them. I understand that removing invalid tags is more efficient than modifying or randomizing them, but I'd like to run a non-reversible learning process.
That was fast, I poked around a bit and put it aside ) I'll finish my version, let's see
degeneracy happens as predicted - interesting. There is a reason to think about how to deal with it.
i read about causal inference the other day, i wanted to use it to search for something... but it seems that it's not our topic
Z.I. got this with a meta model (without iterative learning) in 5 years. Training for 5 months.
Will be thinking how to attach iterative (redoing examples from articles)
That was fast, I poked around a bit and put it aside) I'll finish my version, we'll see.
The degeneration is happening as predicted - interesting. There is a reason to think about how to deal with it.
i read about causal inference the other day, i wanted to use it to search for something... but it seems it's not our topic
Z.I. got this with a meta model (without iterative learning) in 5 years. Training for 5 months.
Will be thinking how to attach iterative (redoing examples from articles)
It looks cool, but I don't understand it yet. If you don't mind telling me what the meta model is or tell me where to read it, maybe you've already written about it in this thread?
I just dropped out of here for a long time, and was not able to follow the thread.
It looks cool, but it is not clear yet. If you do not mind telling me what kind of meth model? Or tell me where to read? Maybe you have already written about it in this thread?
I just dropped out of here for a long time and have not been able to follow the branch.
it's the second model that allows/denies to open trades
i.e. there are 2 models in production