EA31337 released for free | Page 10 | 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

  • #181
  • May 11, 2018 12:38pm May 11, 2018 12:38pm
Quoting kenorb
{quote} Each strategy trade function (Trade_X) got this line: result &= signal_method <= 0 || Convert::ValueToOp(curr_trend) == cmd; so when X_SignalMethod is 0 or positive, it only trades with the current trend. Otherwise, when negative, it trades all the time. The easiest would be to run optimization and find the better TrendMethod which works for the current months, as EA was optimized using older years (and the market is always changing). If you'd like to compile v1.072 (tag), you need to clone the main repo recursively and find the right HEAD...
Thanks,

It sounds a little complicated, maybe I can do it with a little spare time.

Yes, I wasn't expecting the EA to handle a big dd with the current profits which was 70% in less than 3 weeks, which is pretty crazy. Once in a while I have no problem closing trades on a EA due to trend changes or market changes. It is just knowing when to do it. I closed the trades on the smaller account and it is up 42% including losses which nobody can complain about
  • #182
  • Edited 2:11pm May 11, 2018 1:58pm | Edited 2:11pm
Quoting chrisdk2017
{quote} Yes, I wasn't expecting the EA to handle a big dd with the current profits which was 70% in less than 3 weeks, which is pretty crazy. Once in a while I have no problem closing trades on a EA due to trend changes or market changes. It is just knowing when to do it. I closed the trades on the smaller account and it is up 42% including losses which nobody can complain about
I wasn't expecting 70% in the first 3 weeks on the forward testing, that sounds very promising, so thanks for testing. Even with the loses, this still shows a great hidden potential. I hope it's going to continuing like that. I was mainly focusing on the backtests which didn't perform so well and they looked very unstable and didn't know exactly how to approach the problem due to limitations. As far as I remember, v1.072 was heavily optimized (update below), but my understanding is, the latest version of EA has been optimized only to its ~20-30% full potential and also needs some heavy optimization work, that's the reason most of the strategies were disabled by default. This month I can introduce to the method of backtesting using Docker in order to attempt the optimization of the latest version.

Here are the optimization results which were used for the EA, so it can give a brief idea how complex it is:
- v1.072 part 1 (combined results from 1084 commits) + v1.072 part 2 (combined results from 2281 commits)
Assuming each commit approximately is one optimization (~few hours run each) or backtest run (~1h), it took over 3365 tests (the number could be doubled as not all were accepted to be merged, failed or needed to be re-run) to find the right parameters for it. Let's assume each run is around 1h (in reality, some runs took days in a row), it makes over 140 days which is like running optimization tests constantly for over 4.5 months in a row (24h/7).
And here are the results for next versions:
- v1.073 RC1 (1508 commits/tests)
- v1.074 (327 commits/tests)
- v1.075 got only fixes and some manual optimization runs, no further optimizations are done yet.
  • #183
  • May 11, 2018 11:55pm May 11, 2018 11:55pm
Quoting kenorb
{quote} I wasn't expecting 70% in the first 3 weeks on the forward testing, that sounds very promising, so thanks for testing. Even with the loses, this still shows a great hidden potential. I hope it's going to continuing like that. I was mainly focusing on the backtests which didn't perform so well and they looked very unstable and didn't know exactly how to approach the problem due to limitations. As far as I remember, v1.072 was heavily optimized (update below), but my understanding is, the latest version of EA has been optimized only to its...
I tried v1.075 with default settings for 1 week. I only changed trailing stop to 0. But there was only to much trades with bad results overall. Some part of history on picture. Next week I will try v1,072. Can you publish setting which you are using as *.set* file. Please !!!!!!!!
Attached Image (click to enlarge)
Click to Enlarge Name: Snímek obrazovky (122).png Size: 306 KB
  • #184
  • May 12, 2018 1:06am May 12, 2018 1:06am
Please read the whole thread, the set file has been mentioned a number of times.

Last week was not good for the ea, at least 1.072.
  • #185
  • May 12, 2018 2:20am May 12, 2018 2:20am
I looked in the commits for the optimization of v1.074 and CloseOrderAfterXHours is now set to 24 hours rather than 72 hours like in the set for v1.072. I am going to try that for the next weeks of forward testing.
  • #186
  • May 12, 2018 3:58am May 12, 2018 3:58am
Quoting spojka
{quote} I tried v1.075 with default settings for 1 week. I only changed trailing stop to 0. But there was only to much trades with bad results overall. Some part of history on picture. Next week I will try v1,072. Can you publish setting which you are using as *.set* file. Please !!!!!!!! {image}
The EA has been optimized only for EURUSD, it seems you're testing it also for GBPUSD and USDJPY, so I'm not surprised about the losses (looks like stop losses are too tight for GBPUSD). Currently, EURUSD is still the challenge, in the further versions I hope it's going to be optimized for other. So you'll need to optimize params for each pair separately, till then, try or post the results for EURUSD.
  • #187
  • Edited 10:30am May 12, 2018 10:06am | Edited 10:30am
Quoting chrisdk2017
I looked in the commits for the optimization of v1.074 and CloseOrderAfterXHours is now set to 24 hours rather than 72 hours like in the set for v1.072. I am going to try that for the next weeks of forward testing.
The CloseOrderAfterXHours is a very speculative param. On one side, during trading you want to keep your order as long as it's necessary till you hit your stops (e.g., just in case the order gets better, or you get attached to it), on the other hand, during backtesting, you don't want to keep your orders too long to avoid few lucky trades poisoning your overall results.

In general, the optimization results for v1.073 shows, the longer you wait, the better.
Period 30 Minutes (M30) 2014.01.06 00:00 - 2016.12.30 21:30 (2014.01.01 - 2016.12.31)
Model Control points (a very crude method)
Initial deposit 10000.00; Spread 20
Pass Profit Total trades Profit factor Expected Payoff Drawdown $ Drawdown %
1 5963.88 4302 1.04 1.39 12027.22 72.10 CloseOrderAfterXHours=0;
4 4548.52 5783 1.03 0.79 9646.13 65.65 CloseOrderAfterXHours=72;
6 3970.98 5695 1.03 0.70 7842.83 51.07 CloseOrderAfterXHours=120;
3 3285.71 6511 1.02 0.50 10278.19 66.28 CloseOrderAfterXHours=48;
2 1001.18 8141 1.01 0.12 9911.01 70.30 CloseOrderAfterXHours=24;
5 314.52 6091 1.00 0.05 10293.47 71.50 CloseOrderAfterXHours=96;

And another one for 2016 (missing other values probably means they failed to provide positive profit):
Period 30 Minutes (M30) 2016.01.06 03:00 - 2016.12.30 21:30 (2016.01.01 - 2016.12.31)
Pass Profit Total trades Profit factor Expected Payoff Drawdown $ Drawdown %
1 17370.37 1727 1.27 10.06 8451.76 44.77 CloseOrderAfterXHours=96;

So for the forward trading, I think it's probably better to keep it 72h or 0 (for none), for optimization tests, to find the best technical params, it's better to keep it tight (like 24h) as the technical indicator conditions which were used to open the trade are no longer relevant the next day as support, resistance and pivot points changes, so the profit results from the next day should be not considered. On the other side, when you optimized params using 24h closure, you don't know whether you will improve or worse things when you change it during forward trading, because the change will alter the logic which hasn't been tested.

Here is another optimization results for v1.074:

Period 30 Minutes (M30) 2014.01.06 00:00 - 2016.12.30 21:30 (2014.01.01 - 2016.12.31)
Model Every tick (the most precise method based on all available least timeframes)
Initial deposit 10000.00; Spread 20
Pass Profit Total trades Profit factor Expected Payoff Drawdown $ Drawdown %
1 1283.37 779 1.28 1.65 502.41 4.85 CloseOrderAfterXHours=0;
6 1154.21 779 1.25 1.48 502.41 4.80 CloseOrderAfterXHours=120;
5 1117.84 779 1.24 1.43 502.41 4.91 CloseOrderAfterXHours=96;
4 881.71 499 1.26 1.77 396.87 3.75 CloseOrderAfterXHours=72;
2 -54.77 221 0.95 -0.25 338.19 3.36 CloseOrderAfterXHours=24;
3 -54.77 221 0.95 -0.25 338.19 3.36 CloseOrderAfterXHours=48;

So what's the best value for it? I don't know. But it's good to find out on the forward testing.
  • #188
  • May 12, 2018 11:15am May 12, 2018 11:15am
Ok then I change it back.

Do you have a solution for the reversal of trend like we have just seen?

Other than manual closing once in a while?
  • #189
  • May 12, 2018 12:18pm May 12, 2018 12:18pm
Quoting chrisdk2017
Ok then I change it back. Do you have a solution for the reversal of trend like we have just seen? Other than manual closing once in a while?
You don't know when the trend is reversed unless it's too late. Usually, it's done by checking whether the support/resistance is broken, but EA doesn't have such feature. So the only way is to optimize TrendMethod and find the optimal value for the current year or months, so EA will react quicker, or as you say, close the trades manually.
  • #190
  • May 13, 2018 4:58am May 13, 2018 4:58am
Quoting kenorb
{quote} The EA has been optimized only for EURUSD, it seems you're testing it also for GBPUSD and USDJPY, so I'm not surprised about the losses (looks like stop losses are too tight for GBPUSD). Currently, EURUSD is still the challenge, in the further versions I hope it's going to be optimized for other. So you'll need to optimize params for each pair separately, till then, try or post the results for EURUSD.
There are results for EUR USD. Tight SL and extremely tight TP.
Attached Image (click to enlarge)
Click to Enlarge Name: Snímek obrazovky (124).png Size: 204 KB
  • #191
  • May 13, 2018 9:23am May 13, 2018 9:23am
Quoting spojka
{quote} There are results for EUR USD. Tight SL and extremely tight TP. {image}
You've got like ~50/50 wins/losses with minimum lot size, so that's not so bad. Assuming these losses belongs to the specific strategy which doesn't work for the current market conditions, once it hit profit factor below 0.5, they will be automatically disabled, so you'll be left with the one which works. Try for another few days on the same run and share how it goes. The tight SL/TP depending on the strategy params, they usually depend on some other technical indicator values so it may vary.
  • #192
  • May 13, 2018 10:17am May 13, 2018 10:17am
Quoting kenorb
{quote} You don't know when the trend is reversed unless it's too late. Usually, it's done by checking whether the support/resistance is broken, but EA doesn't have such feature. So the only way is to optimize TrendMethod and find the optimal value for the current year or months, so EA will react quicker, or as you say, close the trades manually.
With optimization on Main trend method I get 112 instead of 203. It goes from loss to profit in period 2017-present date. I used Alpari data which is supposed to be better than usual MT4 historical data.

(this is for v1.072)
  • #193
  • May 14, 2018 6:00am May 14, 2018 6:00am
I have tried to optimize GBPUSD using v.1.075.

Very little profits in the last few months, maybe that is just the nature of GBPUSD.

But overall nice profits :-)

from 2k to 48k 2013-present.

Attached File(s)
File Type: txt EA31337-1.075-gbpusd_fully_optimized.txt   33 KB | 437 downloads
  • #194
  • May 15, 2018 12:38am May 15, 2018 12:38am
For CCI, is there a maximum value for the signal method?

It just says "0-" with no maximum in the inputs. I looked through the source and couldn't find the information.
  • #195
  • Edited 4:42am May 15, 2018 4:11am | Edited 4:42am
Quoting chrisdk2017
For CCI, is there a maximum value for the signal method? It just says "0-" with no maximum in the inputs. I looked through the source and couldn't find the information.
CCI was added in the recent version. The range for the signal method is between 0-63 as per the following code (in Trade_CCI()):
if ((signal_method % 1) == 0) result &= cci[period][CURR] > cci[period][PREV]; # 1
if ((signal_method % 2) == 0) result &= cci[period][PREV] > cci[period][FAR]; # 2
if ((signal_method % 4) == 0) result &= cci[period][PREV] < -signal_level; # 3
if ((signal_method % 8) == 0) result &= cci[period][FAR] < -signal_level; # 4
if ((signal_method % 16) == 0) result &= cci[period][CURR] - cci[period][PREV] > cci[period][PREV] - cci[period][FAR]; # 5
if ((signal_method % 32) == 0) result &= cci[period][FAR] > 0; # 6
The conditions are implemented bitwisely (logical AND operation). For example, CCI has 6 main 'if' conditions, so by checking 0-64 range, you can basically test all the permutations. So 0 (for CCI*_SignalMethod) means no filter signal conditions (that is every trade will be executed; high volume trading), '1' value means execute only when 1st condition is matched (CURR>PREV - current CCI value is higher than from the previous bar for buy, opposite for the sell), '2' - execute only on 2nd condition, '3' (1+2) - execute on both 1st and 2nd conditions, '4' - only on 4th conditions, '5' (1+4) only on 1st and 4th, '6' (2+4), '7' (1+2+4), '8', '9' (1+8) and so on, and '63' means all 6 conditions needs to be matched to execute the trades. This is very powerful, as you can test all the combinations for each timeframe and set the unique conditions for each strategy and each timeframe which works for the given market conditions (pair). In the later versions, new conditions can be added or the weakest conditions can be replaced with the better one.

Based on above logic, here are the hints to troubleshoot the volume of trades for the given strategy based on its signal method (filtering system):
- if the strategy trades too often: increase or optimize its corresponding SignalMethod param;
- if the strategy doesn't trade much or you want it to trade more often: decrease or reset SignalMethod to 0, then optimize its value.
  • #196
  • May 15, 2018 4:38am May 15, 2018 4:38am
Thanks for the info.

I think I am getting the hang of the optimization of the EA though the way I optimize is of course much simpler than the months of testing that you have described for the EURUSD set.

I am first running it from 2014-2016 with everything enabled, then optimizing each type of signal (bands etc) and then combining everything to one set finally.

The GBPUSD set took roughly one day of work so I guess you could find even better sets with more optimization.
  • #197
  • Edited 5:27am May 15, 2018 5:03am | Edited 5:27am
Quoting chrisdk2017
I am first running it from 2014-2016 with everything enabled, then optimizing each type of signal (bands etc) and then combining everything to one set finally.
Few hints on optimizations:
- I usually disable boosting (Boosting_Enabled=0) and actions (Account_Conditions_Active=0) when running optimizations, so you've pure technical results. Boosting and actions can be tested at the final test, whether they add some value on top of it or not.
- I also use CloseOrderAfterXHours=48, LotSize set to 0.01 and ProfitFactorMinToTrade/ProfitFactorMaxToTrade=0 specifically for the optimization tests (see ea_default_sets()). So profits for orders opened after 2 days are not taken into the account, and lot size is not automatically increased with the balance, so you're relying on the very strict rules (all trades should be equal as possible, not dependent on the previous success or current balance).
- Above strict rules will not always give you the good results, only the right direction, but combining them with some later-stage params, they can hugely improve.
- Another benefit of setting fixed and minimum lot size (0.01), is that you want to open as many trades as possible (>1000), so you can optimize params based on the large dataset giving as much possible scenarios and different market conditions.
- I usually optimize each strategy on its own, then merge it with the others at the final tests when optimizing params which affects all strategies (see: optimization tests for Strategies).
- SignalMethod should be optimized only for the given strategy and its timeframe (in the isolated run, so every other strategy should be disabled), it's quicker and more reliable.
- If SignalMethod is too much filtered (less than <100 trades), that's sometimes bad, as you don't have enough dataset to optimize other params (in other words, the optimization results could be not reliable).
- Strategy-specific multi-timeframe-shared params such as Period_Ratio, SignalLevel, Applied_Price are tested after optimization of timeframe-specific param (such as SignalMethod).
- For very slow optimization tests, I use Control points to have some rough idea which value is better, but the results could not always be reliable.
- At the final tests, I test boosting on/off or/and actions on/off (or few specific actions), CloseOrderAfterXHours, RiskMargin, RiskRatio, RiskRatioDecreaseMethod/RiskRatioIncreaseMethod and similar.

Quoting chrisdk2017
For CCI, is there a maximum value for the signal method? It just says "0-" with no maximum in the inputs. I looked through the source and couldn't find the information.
Actually, the range for CCI is -63-63, similar for every other strategy (that is, not from 0, but from the negative corresponding value). So as already explained the 6 conditions, negative and positive numbers are basically the same conditions, but positives values trades with trend only, and negative values trades ignores the trend check. So 63 value means all 6 conditions need to match as well the trend need to match the type of order (buy for bullish trend, sell for bearish trend), where -63 will execute trades for all 6 conditions including trades against the trend. The trend logic is controlled by TrendMethod param in v1.072 (based on bear/bull candles), and TrendPeriod in v1.075 (based on RSI).
For example, some strategies aren't really good at trading against the trend (such as Fractals, despite the conditions), so positive values should be set, some other strategies can work well when trading against the trend (like RSI). Including negative values, makes the optimization process twice as long.
  • #198
  • May 15, 2018 9:47pm May 15, 2018 9:47pm
Quoting kenorb
{quote} The compiled version would work the same way. You need to read the error and post it, also make sure you're using Full one, since Backtest is only for backtesting purposes.
Hi,
This was the error encounter with V1.075 BackTest.
There was no error for V1.072
Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot3.png Size: 53 KB
  • #199
  • May 16, 2018 3:57am May 16, 2018 3:57am
Quoting Iry
{quote} Hi, This was the error encounter with V1.075 BackTest. There was no error for V1.072 {image}
There is some initialization bug in the latest version (related posts: #126-127) related to wrong default param values, to fix it, change Envelopes_Deviation_Ratio param to 1.0. Alternatively, you can make Envelopes M15 not active, or set 'Validate startup settings' param to False. This is going to be fixed in the next version.
  • #200
  • May 16, 2018 11:33pm May 16, 2018 11:33pm
Man, your EA is scary AF! What sorcery did you put here?
Note the winrate at 38%, while maintaining a profit factor of 1.98. Superb!
Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot_20180517-152617_1.jpg Size: 108 KB

Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot_20180517-152557_1.jpg Size: 113 KB
1 trader viewing now