Is there a pattern to the chaos? Let's try to find it! Machine learning on the example of a specific sample. - page 26
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
Your method cannot be used directly on my data. Because by training on the previous bar, you are looking a little into the future.
But if you make an embargo section of 1000 lines, you will have a reliable result.
Apply for training not D(i)=D(i-1)+ Target_100_Buy
а
skip the nearest 1000 lines to the current bar. Probably so D(i)=D(i-1000)+ Target(i-999) - but I am not sure. I'll have to think about it. In general, it is necessary to add a shifter for 1000 rows.
P.S. If Alexey's data can also contain several incomplete trades at the same time, then there will also be a peek through the one that has not yet completed, but has already been submitted to the input for training.
Your method cannot be used directly on my data. Because by training on the previous bar, you are looking a little into the future.
But if you make an embargo section of 1000 lines, you will have a reliable result.
Apply for training not D(i)=D(i-1)+ Target_100_Buy
а
skip the next 1000 lines to the current bar.To be honest, I didn't understand the speech at all ... :(
The formula is applied to a row that describes the delta of movement between steps. What kind of peeking into the future?
To be honest, completely misunderstood the match .... :(
The formula is applied to a series that describes the delta of motion between steps. What kind of peeking into the future?
In my sample there is no "between steps" - up to 100 or more steps will be made simultaneously (i.e. trades are not closed, but have already entered the markup, so they should be skipped).
"What did I do?":
The sample train is about 1GB in size. It takes quite a long time to load it into the workspace. I have an i5-3570 with 24GB of RAM and a fast SSD and it takes several minutes for Excel to open this file. That's why I decided that it should be shortened. I was too lazy to figure out the superscripts for 5000+ columns. I took 5584 5586 column and applied a signal to all rows, for example BUY (to be honest, I don't remember which one, maybe SELL). Thus, this column formed a chart according to the above formula. I.e. the first step was zero, then 0.00007, then 0.00007-0.00002=0.00005, then 0.00005+0.00007=0.00012, etc. I.e. from column 5584 5586 I have formed a motion chart with no binding, so to speak a relative motion chart. As if it were a Close chart, i.e. at the end of each step of the chart, the price of the asset changes by the corresponding value.
P.S. Cheated about the column number ... I took the most recent 5586 (I just looked it up in Excel) with the SELL signal.
"... why a new sample":
To show and tell in a certain amount about the approach on its example. If you give me the numbers of columns where OHLC or just Clause prices can be taken, that will be enough.
About the rest:
The data from the sample files is not used at all. On the basis of columns 5584 5586 from each file, a graph is made as described above. And the approach is already applied to these obtained graphs.
Well, since the topikstarter does not want to give new samples, I suggest anyone interested to post their own ... :)
Regards, RomFil!
In Excel the count goes from one, and in CatBoost and mql (and other languages) from zero.
I.e., as I understand, you just took the last column, made an array accumulative, you got a kind of graph. Let's say. You created some predictors based on this data. And the target is the next value of this series, or the original, i.e. delta? I.e. a regression model that gives the result conditionally (+x||-x), and if +x, we enter the trade, right?
I'll try to give the data for those last columns, but a bit later - some changes were made by me to the code since then, then they were lost and everything was reworked again - hard case.
In Excel, the count goes from one, and in CatBoost and mql (and other languages) from zero.
I.e., as I understand, you just took the last column, made an array accumulative, and got a graph. Let's say. You created some predictors based on this data. And the target is the next value of this series, or the original, i.e. delta? I.e. a regression model that gives the result conditionally (+x||-x), and if +x, we enter the trade, right?
I'll try to give the data for those last columns, but a bit later - some changes were made by me to the code since then, then they were lost and everything was reworked again - hard case.
There are no "between steps" in my sample - there will be up to 100 or more simultaneous steps (i.e. trades not closed, but already in the markup).
Still don't get it ... :( What trades, what markup?
The trading approach is as follows:
1) There is a price movement (Close chart, for example bitcoin). A muving with a period of 9 and a shift of -2 is drawn on the chart for clarity.
2) Trading with the approach described above implies signals to sell or buy an asset without being tied to the lot. At one moment of time there is one trade on the asset.
3) If the deal gave a profit, the total is recorded +A number of points, otherwise -A.
4) This is how the income in points is formed.
It is clear that if you add spread and commission to the profit charts mentioned above, the pictures will not be so rosy.
2) Trading with the approach described above implies signals to sell or buy an asset without being bound to a lot. At one moment of time there is one trade on the asset.
In my markup there can be up to 100 or more trades simultaneously. So it makes no sense to apply your algorithm on mine. It will be peeking.
Alexey - can there be several incomplete trades in your data at the same time? I.e. the next signal has appeared, but the deal on the previous signal is not completed yet?
No, in those data there are only consecutive trades.
Still don't get it ... :( What deals, what markings?
Deals according to this markup https://www.mql5.com/ru/code/903
We add 1 trade on each bar and each one waits for its TP or SL. A deal from the previous bar is usually not completed by the beginning of the next bar. In total, there will be many trades at the same time.
No, there are only sequential trades in that data.
Then the RomFil method is not peeking at your data. Not a bad result.