EA31337 released for free | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

No Free Lunch but all the Free Coffee you can drink 736 replies

SWAP free, rollover free Broker? 15 replies

Hi...best time to place order after news released 0 replies

Flyer! Version 2.00 is released. 0 replies

Cad News Released 1 Min Before... ! 43 replies

Attachments: EA31337 released for free
Exit Attachments

EA31337 released for free

  • #21
  • Edited 6:42pm Apr 19, 2018 5:47pm | Edited 6:42pm
Quoting pedrofx
{quote} Backtest always provide very good results. Do you have forward test results? Myfxbook?
Backtests provide good results when you limit your optimization on a fixed dataset (it's kind of play and rewind button, so it doesn't really take into the account what-if you would start at the specific point of time). My approach was to optimize against 2-3 years only (2014-2016), then run the backtests against all years separately using the same parameters, this way you know it works in the new market conditions.

Most of EAs run optimization process on the same fixed period as they do backtesting. So when they run backtest for 5-10 years in a row on the same dataset, they look very smooth. Also, it looks smooth because authors usually set low spreads as 0 or 10 and usually hide that information from the screenshots, or it looks smooth due to the long period of testing and you don't see most of the problems which they're struggling with as they're managing the account with a large unreal sum, secondly tiny change such as a starting date could change the results dramatically, so it doesn't really take into the account what-if you would start at the specific point of time. Therefore I've tests run for 2010,2011,2012,2013,2014,2015,2016 separately each using the same parameters. I'm not claiming it works perfectly, but I've tried my best.

I don't have any 3rd party providers as everything takes time and money to maintain. I had once VPSs where Windows was constantly restarting, or MT4 was disconnecting or updating for unknown reasons, or some other MT4 common anomalies, so I was working on the reliable environment where I can run EA in the isolated environment on Linux for stability reasons, but I didn't have time to finish it.
  • #22
  • Edited Apr 20, 2018 5:16am Apr 19, 2018 7:04pm | Edited Apr 20, 2018 5:16am
Quoting Rafaz1
The least we can do is look into it to see if we can learn from it and help towards our own development/testing/strategy on Forex. Because of the complexity of the EAs it is sensible to back-test on low spread currency pair ranges. So if some of you can do this and report back would be greatly appreciated. Raf
Manual backtesting of EAs is a very complex process, especially this one due to trading on multiple timeframes at the same time, and the results from other people may be not always reliable as around 80% of people who will share the charts, will get it wrong. There are too many variables that could go wrong. Here are few common errors:
- Using Birt's CSV2FXT old script to convert their CSV data into FXT. This script is old, no longer maintained and buggy, so the dataset could be not reliable as you would think.
- Setting FXT files to read-only. What happens when you don't set it? It's going to validated by the platform and most likely wiped out (zeroed) as invalid and your test won't proceed. That means your dataset was never reliable and by setting it read-only, you're forcing platform to accept it. That's why I've written my own tools to convert the data, so the files can be validated (checked for data errors) by the platform itself and accepted when the files have the proper write permissions.
- Using so-called 99% quality data. Such thing doesn't exist. It is a myth sold by scammers and other people still believe it. The 99% modelquality is nothing more than a single, hardcoded number in your FXT file, that's the reason why scam people advise you to set your FXT files read-only, to perform tests on fake 99% quality data which in reality the dataset can be non-reliable not at all, not accepted by the platform, but you're basically forcing the platform to accept it.
- There are many other factors which could go wrong, converting FXT, but missing some HST, or polluting the data with current data of your broker, etc.

Based on above, when you optimize your EA on not reliable data, guess what would happen. There is no evidence that's the results what you see are reliable at all. No matter whether they're smooth or not. In the end, it's just chart on the screen.

For example, when I've found few bugs in scripts which were converting CSV data into HST format miscalculating OHLC data, guess how the charts looked after the optimization. They EA didn't have any struggles to generate smooth charts. After fixing the issues, I had to spend several weeks to re-optimize everything from scratch and generate new reliable charts.

So, most of the backtesting that people perform, usually are based on some old non-reliable steps. Therefore I've created some guide page how to backtest the EA, I'm not claiming it's a perfect method, but given the platform limitations and based on my understanding, it's much more reliable test that other people will perform using old, non-reliable scripts or datasets, which they believe it works without validating the results for them-self.
  • #23
  • Apr 19, 2018 7:12pm Apr 19, 2018 7:12pm
I would like to see longer time frames than 30 mins in the strategies. I also find it problematic to configure how the EA closes trades. A basket TP feature would probably also work well.
  • #24
  • Apr 19, 2018 7:20pm Apr 19, 2018 7:20pm
Any chance to add trading time to this ea? thank you
  • #25
  • Apr 20, 2018 1:41am Apr 20, 2018 1:41am
Quoting kenorb
{quote} Manual backtesting of EAs is a very complex process, especially this one due to trading on multiple timeframes at the same time, and the results from other people may be not always reliable as around 80% of people who will share the charts, will get it wrong. There are too many variables that could go wrong. Here are few common errors: - Using Birt's CSV2FXT old script to convert their CSV data into FXT. This script is old, no longer maintained and buggy,...
You are right, that's the reason why I ask for forward test results? Myfxbook?
  • #26
  • Apr 20, 2018 4:34am Apr 20, 2018 4:34am
Quoting kenorb
{quote} Manual backtesting of EAs is a very complex process, especially this one due to trading on multiple timeframes at the same time, and the results from other people may be not always reliable as around 80% of people who will share the charts, will get it wrong. There are too many variables that could go wrong. Here are few common errors: - Using Birt's CSV2FXT old script to convert their CSV data into FXT. This script is old, no longer maintained and buggy,...
Kenorb,

Thanks for laying out a comprehensive check list...

Could you please confirm if understood this correctly. Are the above pitfalls laid out by you, are overlooked by most if not all? So in general the Test Results for any other EA floating in the forum could potentially be wrong?

Thanks
  • #27
  • Apr 20, 2018 4:49am Apr 20, 2018 4:49am
Hi Kenorb. This EA is an impressive piece of code. Thanks for sharing.

I think I have downloaded all files required but I getting these 2 compiling errors with MT4. Could you please help fixing these two? Thanks

Attached Image (click to enlarge)
Click to Enlarge Name: Image1.png Size: 27 KB
  • #28
  • Apr 20, 2018 5:18am Apr 20, 2018 5:18am
Subs. Thanks and
regards
  • #29
  • Apr 20, 2018 5:44am Apr 20, 2018 5:44am
i tested v1.075 full in EURUSD m15 but didnt got any good result.all losing account . is there any set file? i used default
  • #30
  • Apr 20, 2018 6:10am Apr 20, 2018 6:10am
Quoting TraderTero
I would like to see longer time frames than 30 mins in the strategies. I also find it problematic to configure how the EA closes trades. A basket TP feature would probably also work well.
It's possible to add H1 and more custom timeframes such as M2,M3,M4,M6,M10,M12,M20,H1,H2,H3,H4,H6,H8,H12, but then the params would be too difficult to manage. The EA needs the heavy refactor first before that happens.
  • #31
  • Apr 20, 2018 6:12am Apr 20, 2018 6:12am
Quoting Aimak
Hi Kenorb. This EA is an impressive piece of code. Thanks for sharing. I think I have downloaded all files required but I getting these 2 compiling errors with MT4. Could you please help fixing these two? Thanks {image}
This is the platform bug, you need to upgrade your MetaEditor. That's the thing, every platform version they change something in the language syntax, and code either breaks or not.
  • #32
  • Apr 20, 2018 6:18am Apr 20, 2018 6:18am
Quoting pedrofx
{quote} You are right, that's the reason why I ask for forward test results? Myfxbook?
EA was developed in the isolated environment, the idea was to not trust anything 3rd party, so everything (every utility involved) was developed fully from scratch apart of obvious dependency which is the platform. The MT4 platform doesn't support forward tests, only MT5. EA has partial support for MT5 platform, but it needs some further work. I never run in the long term, because there was always something to improve, so the results would be mixture of different version running on the same account. I wanted to run it somewhere, but it never got into a perfect state where I was happy to run it. I only once run some old version which I don't remember, which from $350 it get into $2200 within few weeks, then forgot about the account and lost it due to some bug which I believe I've fixed it later on. But it's a matter of the right configuration.
  • #33
  • Edited 9:09am Apr 20, 2018 6:54am | Edited 9:09am
Quoting aaven
{quote} Could you please confirm if understood this correctly. Are the above pitfalls laid out by you, are overlooked by most if not all? So, in general, the Test Results for any other EA floating in the forum could potentially be wrong? Thanks
Most of the people who were sending charts which looked very bad, they were performed in the wrong way without being aware of it. So I had to spend a lot of time explaining how to do the backtesting, most of them gave up after that since it was too complex process. Since then I've added many checks to babysit the wrong backtest, but still, you can backtest in the wrong way (old methods, wrong converted dataset, see post #22).
One of the checks is to make sure all indicators are initialized properly and not returning any rubbish, post #5). Secondly, any performed backtest is also accompanied with the text file (generated by EA at the end of the run) such as this one (also attached). So so-called 99% quality test, could be the very worst.
Things to note during good backtesting is Ticks per min and bar, Total ticks and bars.

- If 'Ticks per min' are lower than 60, then it's like one tick per second, so it's not a really good simulation. Good tick data has few hundred of ticks per minute. This relates to your FXT data.
- If 'Total ticks' doesn't sum up, like one tick or less per second for the whole period, that means the dataset is rubbish or consist huge data gaps. For example, large price gaps are reported at the bottom of the file, however, still the price gaps are not the same as data gaps.
- Normally, a year test should have at least few hundred millions of ticks. For example, for 2014 year I have over 73mlns ticks, for 2015 - 80mln. So if you run backtest for multiple years with less than 100mln of ticks, there is something wrong with your data.
- 'Total bars' needs to sum up to what you're testing as well. This relates to your HST data.
- The general rule is, if ticks are not right, check your FXT data, if bars are not right, check your HST files.
- Furthermore, the report file shows the calculated real 'Modelling quality' (see: CalcModellingQuality()), not the fake/hardcoded one, and it's calculated based on the official docs from MT4. This value is more like bar quality between timeframes, nothing to do with your tick data as the percentage changes based on the timeframe you run. In the end, it still doesn't tell you whether you tick data is good or not. The more important is to how many ticks and bars have been processed during the period.
- Hint: Never use Control points, at least for this EA, always All ticks, as the results can be completely different.

So if somebody post some chart without sharing how the test was performed, such as real modelling quality, total bars, ticks, spread, period, how HST/FXT were generated, whether files were read-only or not, input params, it may suggest that was performed in the wrong way. In this case, I'd request sharing the report file which gives more info how the test was done.

Another way to test that the dataset is reliable is to do:
- Convert CSV into HST and FXT using a script (convert_csv_to_mt.py). Only 1 or 2 days is enough.
- Consider checking HST and FXT data using the script to read binary data back to CSV (convert_mt_to_csv.py) and compare.
- Run EA with the option to export all ticks back into CSV (RecordTicksToCSV) on this short period (1 or 2 max).
- Compare original CSV data with the one generated by EA.
Last time I've checked and the data was 100% accurate. However, this only applies to ticks data and it doesn't validate consistency of HST data (technical indicator data). For example, miscalculation of OHLCV values in HST files could produce different results. I don't have scripts to validate HST, but not setting HST files to read-only, you're allowing the platform to validate it for you.
Attached File(s)
File Type: txt EURUSD-2000USD-2015.01.06-2015.12.30-20spread-(5)-M30-report.txt   4 KB | 529 downloads
  • #34
  • Apr 20, 2018 7:02am Apr 20, 2018 7:02am
Quoting kenorb
{quote} Most of the people who were sending charts which looked very bad, they were performed in the wrong way without being aware of it. So I had to spend a lot of time explaining how to do the backtesting, most of them gave up after that since it was too complex process. Since then I've added many checks to babysit the wrong backtest, but still, you can backtest in the wrong way (old methods, wrong converted dataset, see post #22). One of the checks is to make sure all indicators are initialized properly and not returning any rubbish, post #5)....
Thank you so much Kenorb...

I am still green/Newbie with regards to EA/Testing etc..., but you have saved us in going in the wrong direction...

Much appreciated brother...
  • #35
  • Edited 9:55am Apr 20, 2018 7:13am | Edited 9:55am
Quoting 5aztv88
i tested v1.075 full in EURUSD m15 but didn't got any good result.all losing account . is there any set file? i used default
No set file for the latest version. The latest SET file generated by optimization bots has been translated into the default settings and it's part of the release file.

Basically, EA learns on losses and improves over time, but it can take days or weeks. The challenge is to understand the cause of the losses, whether they're just temporary, or long-term, which is not easy. By default, EA got 21 days to warm up (InitNoOfDaysToWarmUp) to allow minimal losses. During that time it prioritizes which strategy is better than the other, so it manages and prioritizes them according to the settings (such as boosting) so it slowly increases lot size for the strategies which were proven to work in the current market conditions or disable (temporary or permanently) certain group of strategies which doesn't work at all (ProfitFactorMinToTrade). But this process can take time and cannot be guaranteed to work without spending some effort into the proper analysis what's going on.
  • #36
  • Apr 20, 2018 10:42am Apr 20, 2018 10:42am
Hi Kenorb and everyone. Today's forward test on demo looks very encouraging. Results and set file attached.

Attached File(s)
File Type: xlsx EA31337-Lite-Full-v1.072_statement.xlsx   23 KB | 874 downloads

Attached File(s)
File Type: txt EA31337-eu-set.txt   7 KB | 966 downloads


50000 Demo account with 1:200 Leverage
Number of trades: 99
Total lots traded: 150.16
Pips made: 354.8
Profit in $: 3579.24

A few cosmetic suggestions for the EA

  1. Order Comments
  2. Close all button
  3. Profit/Loss and Pip Counter

  • #37
  • Apr 20, 2018 11:02am Apr 20, 2018 11:02am
Quoting shiva
A few cosmetic suggestions for the: EA Order Comments, Close all button Profit/Loss and Pip Counter
- EA Order Comments

The comments are implemented (see order_comment in ExecuteOrder()), I'm not sure why they won't work, could be a platform or EA bug. The comment should contain which strategy opened the position, on which timeframe and the spread. When the position is closed manually by EA, it should also include the reason for closing.

- Close all button Profit/Loss

There are conditions and actions for closing all profits or loss (like 'Close all in profit', or 'Close all!'). The missing part is to have the account condition on EA exit. I've added a new feature request.

- Profit/Loss and Pip Counter

All stats are calculated and shown in the chart (profit/loss pips per hour), further stats are generated in the text file which is generated in files/ folder. Check the file, and if anything is missing, please elaborate your idea.
  • #38
  • Apr 20, 2018 11:06am Apr 20, 2018 11:06am
Quoting shiva
Today's forward test on demo looks very encouraging.
Thanks for doing this. Are you only testing v1.072, or v1.075 as well? Each version is progressing with fixes and logic improvements, however, after the logic changes, the optimized settings could be behind. So I was curious whether you found v1.072 better optimized than v1.075
  • #39
  • Apr 20, 2018 11:07am Apr 20, 2018 11:07am
Quoting kenorb
{quote} - EA Order Comments The comments are implemented (see order_comment in ExecuteOrder()), I'm not sure why they won't work, could be a platform or EA bug. The comment should contain which strategy opened the position, on which timeframe and the spread. When the position is closed manually by EA, it should also include the reason for closing. - Close all button Profit/Loss There are conditions and actions for closing all profits...
Sorry for not being more clear. My thought was that, if the user sees a satisfactory running profit even before the preset targets are met, he/she could click a button to close all trades and retire for the day/week etc. This is what I did - Used an external "close all" script to close all open positions and disable the EA.
  • #40
  • Apr 20, 2018 11:09am Apr 20, 2018 11:09am
Quoting kenorb
{quote} Thanks for doing this. Are you only testing v1.072, or v1.075 as well? Each version is progressing with fixes and logic improvements, however, after the logic changes, the optimized settings could be behind. So I was curious whether you found v1.072 better optimized than v1.075
Only 1.072. Because somewhere in this thread, i read that this version is functioning without issues, so, I chose to run it. Next week I'll run test on both simultaneously.
0 traders viewing now