Machine learning in trading: theory, models, practice and algo-trading - page 65
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 will try to make it shorter:
The idea of Mihail Marchukajtes is not to use the classifier to predict the future price direction, but to take some TA signal and classify its readings for "lying" or "not lying". I.e. the classifier can be used as a lie detector (polygraph) for some signal system based on TA.
I got that. Here "lying" means that the price will not go 100 pips or more before the next signal.
This is also an option, I agree.
Once again, do not confuse the forecast with the clasification. The forecast says that in 5 hours the rate will be 100 points higher or 121 points or 122, and the qualifier says that the current situation suggests that the rate will go up. Let's not confuse the concepts...
I reread it several times, but I can't understand anything.
Especially confusing is the example of the dash. The point is that the same crossing of the wand - the fast one crosses the slow one from the bottom upwards, in the future it may cause the price to rise by the required 100 points as well as to fall.
Yes, but in the first case momentum was 100 and stochastic was 30, and in the other case at the same intersection stochastic was 99 and stochastic was 50. In the first case the signal was true and in the other false, but both signals led to the profit, because if the classifier gives out a false signal, then it goes in the opposite direction from the signal. My picture shows everything ....
But the crossover is a fait accompli, the notorious points based on future movements are not clear. That is, you analyze each bar on whether it will grow after it by 100 points or not.... It's also quite acceptable, but it loads the classifier much, because it has to analyze network of market, and it will be difficult to fit 5-6 months on the clock...
Anyway, you can send me the ksv file, I will train it and send you the binary model, and you check it on your own. How well does it work? But it would be more correct to classify the signals of any system. For example the crossing of the wands.... So it will be more correct when working with the classifier. I think so.... Event occurred. In contrast to your event, the waving gets a face in the form of buy and sell, and your event does not have this face. Just an event... nothing more....
That is, you analyze each bar to see if it will grow after it by 100 points or not.... This is also quite acceptable, but it loads the classifier very much, because it has to analyze the market network, and it would be hard to fit 5-6 months on the clock into it...
The classifier can not predict here you are wrong. The classifier determines the current state of the system, and having determined it we make a conclusion about growth or decline.....
Sorry, I wanted to correct you, not to make you laugh.
But you know better.
The subject of terminology is closed.
I reread it several times, but I can't understand anything.
Especially confusing is the example of the dash. The point is that the same crossing of a wave - the fast one crosses the slow one from the bottom to the top, in the future it can lead both to the growth of price by the sought 100 pips and to its fall
If the classifier, which is used as the "lie detector" of the wands, informs that the wands "do not lie", we open a deal on the wands' readings. If the classifier reported that the wipers are "lying", then we can open a trade in the opposite direction to the wipers' readings.
This was for binary classifiers.
A ternary classifier tells with a "-" sign that it cannot report with adequate probability whether the treadmills are lying or not, thus calling to sit down on the fence and smoke bamboo until the next signal - a treadmill crossing.
Yes, but in the first case momentum was 100 and stochastic was 30, and in the other case at the same crossing stochastic was 99 and stochastic was 50. In the first case the signal was true and in the other false, but both signals led to the profit, because if the classifier gives out a false signal, then it goes in the opposite direction from the signal. My picture shows everything ....
But the crossover is a fait accompli, the notorious points based on future movements are not clear. That is, you analyze each bar on whether it will grow after it by 100 points or not.... This is also quite acceptable, but it is a heavy load for classifier, because it has to analyze network of market, and it would be hard to fit 5-6 months on the clock...
If you knew how annoying the discussion is, when it turns out that you also have to take into account the pockets!
Could you please post the excel file. I would build other models on your data.
No, you haven't thought it through. I can cut entry points where the market will really go up or down by 100 points in the future, and not where some kind of crossover has appeared. In other words, I push everything, while the wave crosses somehow, and I have to classify their signals. I do not understand the usefulness of the signal system.