Machine learning in trading: theory, models, practice and algo-trading - page 1936
![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
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
Anyway, my friends, patterns do exist, you just need to be able to see them, and algorithms can see them better than humans.)
I don't think it's an accident...
This is from Expert, neuro bitcoin forecast for the last week, the first screen without filter, the second is the same, but with a filter of "confidence" 40.
The euro does not give such a quality, but close to it.
When will you trade?
When are you going to trade?
I can't reach it yet, I hit a wall.
65% ?
65% ?
Right now the forecast quality on EVERY candle is about 52-53%. That's on EVERY candle. There are at least three ways to forge more or less good answers that reach 65%, they are few and far between and it's only on the test. If the real market was 65% I would beat the binary, but so far I'm only "fighting back", i.e. it's 56-58% on the real market.
Maybe this is the global limit of my method, maybe not )). Working...
It's not like that...
Have you ever tried to improve input quality instead of forecast quality...
For example to find all spots where the network has more than 60% confidence and create a separate dataset from these spots and train the network again...
If you can't improve the quality of prediction, try to improve the quality of input... and you don't need 80% of error there...
if good input gives you 1 to 6 stop/profit for example you can be wrong 75% of the time and make good profit
Have you ever tried to improve not the quality of the forecast, but the quality of the input...
For example to select all sections, where the network has a confidence > 60% and create a separate dataset from these sections and train the network once again...
If you can't improve the prediction quality, try to increase the input quality... and you don't need 80% of error there...
if good input gives you 1 to 6 stop/profit for example you can be wrong 75% of the time and make good profit
I ask the network question "up or down" on every candle. As soon as I start giving not everything, but only a particular one, the quality drops immediately. I think this is because the network stops learning the real market, it learns our view of the market, and everything that is ours is always wrong ))
I appreciate those 52-53% because those are the patterns that the network itself pulled out and what's in its head is a complete dunce.
Have you ever tried to improve not the quality of the forecast, but the quality of the input...
For example to select all sections, where the network has a confidence > 60% and create a separate dataset from these sections and train the network once again...
If you can't improve the prediction quality, try to increase the input quality... and you don't need 80% of error there...
if good input gives you for example 1 to 6 stop/profit you can be wrong 75% of the time and make good profit
This is a hell of a job, these >60 are obtained on the test site. I.e. you have to turn a test one into a training one + a new test one + the dataset itself will be insufficient for training. I'm not ready for such feats.
This is a hell of a job, these >60 are obtained on the test section. I.e. you have to turn a test one into a training one + a new test one + the dataset itself will be insufficient for training. ... I'm not ready for such feats.
it's a 6 min R job, or rather it's not a job at all, I have no idea why you like python so much
1. Fuck Alexey, if every teacher in your town had an individual perception, one would have the letter "B" as a "zu" and the other as a "69" (individuality). and the other had a "69" (the same individuality), you realize that you still wouldn't be able to read!!! I read what you say, I do not understand, I get nervous, you do not get answers to your questions because I do not understand them, time is wasted, no use and who got better from this idiotic individuality of perception????
Let's calm down and not be nervous!
I agree that synonyms are fiches\predictors/attributes.
But as for understanding the concept of "rule", it's not so unambiguous. A rule is an algorithm of actions that can transform information (leading it to meanings - all sorts of formulas) and systematize it (the same decision trees).
In its bare form, we have the OHLC price, which is the observation of the behavior of the value (the relative estimated value) of an asset. From the price, by applying different transformation rules (arithmetic and logical) we get the result of applying the rules in the form of fics/predictors/signs. Yes, without transformation prices can also be signs - there is no contradiction here.
Dimensionality reduction is a secondary transformation of price through intermediate predictor/attributes. In order to know how they were transformed, it is necessary to know the resulting rules of their transformation, which can be either arithmetic expressions or logical ones. As a result of their transformation, new (additional/alternative) fiches/predictors/attributes appeared.
As a result, in this context, the rule is a description of the process of converting price into fics/predictors/signatures and from the latter into new fics/predictors/signatures.