Nebula start editon for MT4
- 专家
- Vladimir Mikhailov
- 版本: 3.32
- 更新: 20 十一月 2017
- 激活: 10
This Expert Advisor is based on two moving averages. The EA implements a self-training mode allowing it to immediately start trading. However, it is also possible to conduct a preliminary EA training for more efficient trading. By default, the EA is configured to work on EURUSD H1.
The robot is based on the modified probabilistic neural network. Initially, the neural network does not contain neurons. In this mode, the EA only learns. No trading is performed. In the course of the EA's work, it learns, and neurons responsible for one of the two classes (buy or sell) start forming in the neural network.
As soon as there is at least one neuron, the EA can trade.
Settings
- Target type - type of the target level:
- Points - in points.
- Money - in money.
- TargetProfit, TargetLoss, TrailingStop - set stop loss, take profit and trailing stop levels in points. Applied to the symbol as a whole.
- VolumeLot - trade lot volume.
- Slippage - slippage in points.
- Pause after trades, msec - pause after a trade in milliseconds.
- Number of trade attempt - number of attempts to enter the market.
- Indicator type - indicator type:
- Moving_Average - moving average;
- Adaptive_Moving_Average - adaptive moving average.
- Indicator first period - first moving average period.
- Indicator second period - second moving average period.
- Neuron activation level - neurons activation level. This is the minimum value of the network output, which affects the decision making. The recommended value is at least 0.95.
- Neural network signal duration, bars - neural network signal duration in bars.
- Neural network renewal period, days - neural network update period in days.
- Enable load neural network - load neural network data from the file.
- Prefix file name - set the file name for storing the neural network data. The file is created in the terminal's common folder in the Files\NNS directory.
- Enable Info panel - show the info panel displaying data on a neural network and trading levels.
- Expert Magic number - unique identifier of the EA.
- Order comment - order comment.
Preliminary training algorithm and the EA settings
-
Set the zero value of the Target profit, Target loss and Trailing Stop parameters. Set Enable info panel - no. Set the file name in Prefix file name.
Set the Indicator first period and Indicator second period parameters from 1 to 100 or higher. Set the 'Neural network signal duration, bars' parameter value from 0 to 10 or more.
Select the testing and optimization periods, for example, the year of 2016 – for optimization period and 2017 – for the testing one. Launch optimization on the year of 2016.
-
Once optimization is complete, select the appropriate results. When selecting results, adhere to the following recommendations: Indicator first period should be less than Indicator second period. Set other parameters at your discretion.
Run a single EA test with selected parameters on the period of 2016. If the results are suitable, check the EA on the test period of 2017.
-
Run a single test on the test period of 2017. If you are satisfied with the test results, then it is time to use the EA in trading. Set the Target profit level, Target loss level and Trailing stop level parameters at your discretion.
this EA deserves definitely my rating *****5 ,
I have not foundhere another AI honest expert adviser which does not make a mystery about itself by hiding most of its paramaters , sometimes you wonder if these EAs really use NN...
main problem seems to be the not precise enough explanations of the files and how they are attached to the chart ,
the EA uses a config file which is not the same as the normal MT4 setup file. The interaction of these two files is not explained.
Like if the file network.config has to be copied or moved from the tester folder to the mql/files/NNS folder or not...
as ist looks like the first time the file is created automaticallly also in the MQL5/files/NNS folder ( not only in the testers folder). But from the second chart on and for a different pair , one has to do the copying manually ...
I suspect also that the historical data from Metatrader are not good enough for training the net .
A neural net can be difficult to master , but this EA is highly appreciated .
Like any NN it can be trained on nearly any time series , so might try your luck on any pair and any time frame.
This leaves plenty of room for experiments , although the main data to optimize the net are very few , only 3, namely 2 moving averages and a a duration of NNsignal validity .
You might be surprised to find out that a net trained on one pair is giving good results on another pair , or even on another timeframe...
I would hope for maybe an update or even a slightly improved edition , allowing exits based on time, limitation of postions ( MAX open or close ) , exits based on NN sensitivity levels ( which is standard with Neural Nets) etc ... and a telegram group for the users.