My Trading

Instructions and recommendations for using the Neuro Future indicator

5 September 2025, 16:38
Sergey Rozhnov
Sergey Rozhnov
0
493

Instructions for using the Neuro Future indicator.

Content:

  1. Quick Start: Get Started in 5 Minutes
  2. Training a Neural Network: A Detailed Guide
  3. Validation and overfitting control system
  4. Advanced settings
  5. Interpretation of results
  6. Integration with advisors (EA)
  7. Frequently Asked Questions (FAQ)
  8. Recommendations for use
  9. Support

1. Quick Start: Get Started in 5 Minutes

1.1 Installation and initial setup

  • Find "Neuro Future" in the MetaTrader 5 Navigator.
  • Drag it to the desired trading chart.
  • In the settings that open, leave the default values (or select Network type).
  • Set Max training epochs : 2000-5000, Max training samples :1000-2000
  • It is recommended to enable validation to protect against overfitting:
    • Validation period : 200
    • Use validation criteria for early stopping : true
    • Save best weights : true
  • Click "OK".

1.2. First training and forecast

  • On the chart, click the button NEW LRN.
  • Follow the process in the "Experts" journal. The training will be completed automatically.
  • Done! The indicator will go into forecasting mode. The lines on the chart will show future values.

2. Training a Neural Network: A Detailed Guide

2.1 Preparing for training: Choosing a strategy and architecture
  • Selecting the network type (Network type): Determines the logic of the indicator (T1 - price forecast, T1Dif - change forecast, T3Bin - binary classification of the trend, etc.).
  • Architecture setup (L1, L2, L3, L4): Determines the size and complexity of the neural network.
  • L1 (Input layer): The size of the window of prices to be analyzed.
  • L2 and L3 (Hidden Layers): Responsible for the complexity of computations.
  • L4 (Output layer): Number of bars to forecast.
2.2. Launching and managing training
  • NEW LRN: Completely discards the previous network and starts training from scratch.
  • ADD LRN: Continues training an existing network, e.g. on new data (useful for adaptation), or after stopping.
  • STOP: Forces a stop to the tutorial, keeping the current progress.
2.3. Process monitoring and log interpretation
  • The Experts tab displays key metrics in real time (display is configurable by parameters):
  • Epoch - the current epoch number.
  • Error - error on the training sample (tends to decrease).
  • Valid (NetErr, Profit, Binar) - error or criterion on validation data (main quality indicator).
  • The vertical lines on the chart show the range of historical data used for training.

3. Validation and retraining control system

3.1. How validation works
  • The indicator uses a built-in validation system to prevent overfitting:
  • ValidationPeriod - defines the number of the last bars that are not involved in training
  • This data is used solely to check the quality of the model.
  • The EarlyStopping process monitors the validation data for errors
3.2. Recommendations for setting up validation   (10-20% of the number of examples is recommended)
  • For daily timeframes (D1): set ValidationPeriod = 100-200 bars
  • For hourly timeframes (H1): set ValidationPeriod = 300-500 bars
  • For minute timeframes (M15-M30): set ValidationPeriod = 500-1000 bars
3.3 Interpretation of validation results
  • A high criterion (Profit, Binar) or low error (NetError) on the validation data indicates good generalization ability of the network
  • A decreasing criterion (or increasing error) on validation as the training error decreases is a sign of overfitting.
  • The indicator automatically saves weights with the smallest validation error (with the SaveBestWeights option enabled)

4. Advanced settings

4.1 Key learning parameters
  • MaxEpochs - maximum number of training epochs
  • TargetError - target error value for early termination
  • LearningRate - learning rate
  • UseAdaptiveLR - Use adaptive learning rate
4.2. Additional parameters
  • EarlyStopping - the number of epochs without improvements to activate early stopping
  • SaveBestWeights - save the best weights based on the validation error (or if validation is disabled, based on the network error)

4.3. Activation and scaling settings

  • ActivationPreset - preset configurations of activation functions (Auto/Manual)
  • ActivationTypeHidden - activation function for hidden layers (when configured manually)
  • ActivationTypeOut - activation function for the output layer (when configured manually)
  • InputScale - input data scaling method (S11/S01)
  • OutputScale - output data scaling method (S11/S01)
  • GradientLimiting - Enable gradient limiting
  • max_grad - maximum gradient value (with limitation enabled)

4.4 Notification and Logging Settings

  • EnableAlerts - Enable trading alerts
  • AlertThreshold - alert trigger threshold
  • PushNotifications - sending push notifications
  • EmailAlerts - Sending email alerts
  • SoundAlerts - Sound Alerts
  • EnableLogging - enabling the logging system
  • ReduceLog - frequency of logging (reduction)
  • LogExamples - logging training examples
  • LogResults - logging of training results
  • LogLoad - logging network loading
  • LogSave - logging of network saving

4.5. Additional indicator settings

  • UniverseOutputScale - universal output scaling
  • FixIndicatorWindowMinMax - fixing the minimum/maximum of the indicator window
  • MaxBars - maximum number of bars in the indicator window
  • AutoColor - automatic color scheme
  • Color - select color (when AutoColor is disabled)

5. Interpretation of results

5.1. Information panel (GUI)

  • The information panel displays:
  • Network structure - layer configuration (L1, L2, L3, L4)
  • Accuracy - current assessment of the accuracy of forecasts
  • Training period - the time range of data on which the network was trained
  • Activations - activation functions used for hidden and output layers
  • Scale type - the method of scaling input and output data (S01 [0,1] or S11 [-1,1])

5.2. Visual elements on the chart

  • Forecast Line - a colored line that displays the forecast for the selected bar
  • Graphic objects - visualization of future forecasts directly on the price chart
  • Vertical lines - represent the period of data used to train the last loaded network
  • Color indication - informs about the compatibility of the loaded network with the current symbol and timeframe

6. Integration with advisors (EA)

6.1 To call the indicator from the advisor, use the iCustom() function.

  • Example of initialization in the advisor:

int OnInit()
{
   // Loading the indicator 
   indicator_handle = iCustom(_Symbol, _Period, Indicator_Name, FutureBar, File_Name, 0);
   if(indicator_handle == INVALID_HANDLE)
   {
      Alert("Error loading the indicator: ", GetLastError());
      return(INIT_FAILED);
   }   
   return(INIT_SUCCEEDED);
}

6.2 Parameters for optimization in the strategy tester:

  • Prediction Number (1 to 6)
  • Threshold values for generating trading signals (SignalLimit)
  • Network type (Versions) - T1, T2, T3, T4 and their modifications
  • Sizes of neural network layers (LL1, LL2, LL3, LL4)

7. Frequently Asked Questions (FAQ)

Q: The network does not load or does not start training.

A:

  • Check the write permissions in the MQL5/Files/ folder
  • Make sure there is enough historical data available
  • Check the correctness of the specified network parameters (layer sizes)
  • Make sure the network files exist and are not corrupted.

Q: What Network type should I choose?

A:

  • T1 - Basic option. It is recommended to start with it
  • T1Dif, T2Dif - Strategies that analyze price differences. Can be more accurate for determining directional movements
  • T2 - Context-dependent analysis. Takes into account volatility
  • T3/T4 - Specialized strategies for accurate determination of trends and impulses

Q: How to determine the input/output scale type?

A:

  • Check the Type parameter on the indicator information panel (GUI)
  • If the UniverseOutputScale parameter = true, the display in the indicator window is standardized to the range [-1,1]
  • If UniverseOutputScale = false, the output values correspond to the original scale of the selected strategy (S01 or S11)

Q: Why does the indicator use this particular validation method?

A:

  • This approach is standard in machine learning and provides a fair assessment of the quality of the model on data that was not used in training.

Q: How often should the network be retrained?

A:

  • It is recommended to retrain the network whenever market conditions change significantly or every 1-2 weeks to keep the model relevant.

8. Recommendations for use

  • Determine the scale type - Understanding the scale of the output data (S01 or S11) is critical to properly interpreting the signals.
  • Set up thresholds - Optimize the SignalLimit parameter to your trading strategy and selected timeframe
  • Test different types of networks - Strategies based on price differences (T1Dif, T2Dif) can show better results on volatile instruments
  • Consider the time frame - High time frames (H4, D1) often require more conservative (larger) thresholds to filter out noise
  • Periodic retraining - Regularly retrain the network on new data to keep the model up to date
Important validation notes:

  • The validation period is cut off from the end of the historical data.
  • For maximum relevance, it is recommended to periodically retrain the network on new data.
  • The validation period size should match your trading horizon.

9. Support

If you have any questions or problems:

  • First of all, check the logs in the "Experts" and "Journal" tabs. Make sure that logging is enabled in the settings
  • Make sure you have enough historical data for the symbol and timeframe you choose.
  • Determine the type of network used and the data output scale - this information is often needed for diagnostics
  • For complex questions, please contact the indicator's discussion section on the Marketplace or the developer via private messages