Kohonen and Patterns - page 3
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
Do you have a NeuroScalp program? If so, please send it to mike_02.89@mail.ru
I FOUND IT IN THE FREE. I HAVE FOUND IT - BUT IT IS NOT DOWNLOADED ANYWHERE !!! ((
I was looking for one too at one time. I don't think anyone will give it away so easily (I'm even sure). And on sale free it has long been gone. Demo version bitten off to the limit and can not use it really and break it too makes no sense. Also interested in if there is someone will share and for what?
In general, Kohonen Maps allow you to find patterns, and thanks to these maps you can determine whether an indicator is suitable for buy and sell signals.
I tried to do it with some indices... I tried to do it with some indices and the map showed that one group of indices shows buy signals are close to sell, but sells are scattered, so the net shows that these indices are good for buy but not for shorts.
In general, it is interesting to experiment with cards. In particular Kohonen Maps are well suited for integration with technical analysis, i.e. MTS may be run using the maps. (I am preparing an article about these maps).
As for training a simple Expert Advisor Al: I should also look at the number of trades.
Because the number of trades represents the number of patterns (patterns and situations in the market) that the network knows.
I was looking for one too at one time. I don't think anyone will give it away so easily (I'm even sure). And on sale free it has long been gone. Demo version bitten off to bits and can not use it really and break it too makes no sense. Also interested in if there is someone to share and how much?
Do you have a demo version?
If yes, could you send it to my mailbox?
Do you have a demo version?
If yes, could you send it to my mailbox ?
Unfortunately I didn't keep it. Spun it and threw it away. Because it's a very cut down one. I think I downloaded it on a spider.
Unfortunately, I didn't keep it. Spun it and threw it away. Because it's very truncated. I think I downloaded it on a spider.
Thank you very much!
It's a shame about the cut, but at least it's something.
Alas, it doesn't work on the spider either.
As far as I understand -- sliding window training?
Hmm, how do you look at a rather simple (by code, not by result) estimation of the reliability of the result?
I tried, after extrapolation to history, to calculate R^2 coefficient of determination (that's if you're familiar with it, like correlation coefficient, but better satisfying the reliability estimate) between the prediction and the real data. Well he was fully describing what is already apparent to the eye. You have to set the purpose of the forecast and the method of working with it. If you make a rough estimation, for example, if the forecast at the end of the next day is higher than the current one as if it were "Buy", if it is lower - "Sell" and calculate the result, then in general it is more or less the year. During the last year on the 1st lot 3700 points. By the way, I made it so that it is counted automatically, on the first page on 15-minute chart in the comments is Sum=2084, and N=12 - days, that is at this rough estimate 2084$ for 12 days. But this is naturally a very crude and primitive estimate. I am more attracted to the fact that the network is not bad at showing amplitude points, but rather the nature of the change. Perhaps if you work on filtering the trend component you will see reversal points in the middle of the day.
I tried to calculate R^2 coefficient of determination after extrapolating to history...
How complicated it is... and simple at the same time...
How complicated... and simple at the same time...
Hey! You better tell me a little bit about how you're doing?
I tried to calculate R^2 coefficient of determination (if you are familiar with it, like correlation coefficient, but better satisfying confidence estimation) between forecast and real data after extrapolation on history. Well he was fully describing what is already visible to the eye. Here you should define the purpose of the forecast and the method of working with it. If you make a rough estimate, for example, if the forecast at the end of the next day is higher than the current one as if it were "Buy", if it is lower - "Sell" and calculate the result, then in general the year is more or less. During the last year on the 1st lot 3700 points. By the way, I made it so that it is considered automatically, on the first page on 15-minute chart in the comments is Sum=2084, and N=12 - days, that is, at this rough estimate 2084$ for 12 days. But this is naturally a very rough and primitive estimate. I am more attracted to the fact that the network is not bad at showing amplitude points, but rather the nature of the change. Maybe if we work on filtering the trend component we will be able to see reversal points in the middle of the day.
I do not understand you completely. However, I have not completely stated my thought, so I am correcting my mistake.
Suppose we have a network. There is an TS that trades according to the network prediction. The net predicts the days for 5 days ahead.
So. To increase effectiveness of TS you may introduce estimation of forecast, not on history, but in real trading. This estimation can be used as a basis for MM, etc.