Parameters of the “Neuro Future” MT5 indicator with a detailed description

Parameters of the “Neuro Future” MT5 indicator with a detailed description

Content:

I. Neuro Future indicator parameters, default values.

II. Detailed description of parameters.

III. Technical description of the types of networks used.

I. Neuro Future indicator parameters, default values:

1. —— BASIC SETTINGS ——

• Predicting number : 6 – Predicting N-1 bars for T, TBin, and N bars for TDif.
• Network filename prefix   : “net_” – Network file name prefix
• Filename postfix type   : T_ACT_SYMBOL_TF – File name postfix type (symbol+timeframe)
• Use common folder for files   : true – Use shared folder for files
• Auto color by type   : true – Automatic color by type
• Manual color   : clrGray – Manual color selection (if AutoColor is false)

—— NETWORK STRUCTURE ——

• Network type   : T1 – Network Type
• Activation preset   : Auto – Preset of activation functions
• Loss function type   : MSE – Loss Function
• Input layer size   : 40 – Input layer size
• Hidden layer 1 size   : 27 – Size of the first hidden layer
• Hidden layer 2 size   : 12 – Size of the second hidden layer (0 = disabled)
• Output layer size   : 6 – Output layer size (from 2, from 1 for TDif)
• Network comment   : “” – Comment on the network

—— MANUAL ACTIVATION SETTINGS ——

• Hidden layer activation   : Tanh – Activate Hidden Layers
• Output layer activation   : Tanh – Output layer activation
• Input scaling   : S11 – Input Scaling
• Output scaling   : S11 – Output Scaling
• Gradient limiting   : false – Gradient limit
• Max gradient value   : 0.1 – Maximum gradient value

—— TRAINING PARAMETERS ——

• Start learning after loading   : false – Start training after loading
• Use training date   : false – Use training date
• Training date   : DATE – Date of training
• Max training epochs   : 500 – Max number of training epochs
• Max training samples   : 100 – Max. number of examples for training
• Bars between samples   : 1 – Minimum bar between samples
• Use adaptive learning rate   : true – Use adaptive learning rate
• Learning rate   : 0.1 – Learning speed

• Target error   : 0.0001 – Target Error

—— VALIDATION SETTINGS ——

• Validation period   : 0 – Validation period (bars)
• Use validation criteria   : false – Use validation criteria for early stopping
• Validation mode   : Profit – Validation mode
• Validate only selected bar   : false – Validate only ‘Prediction number’
• Validation threshold   : 0.1 – Prediction threshold for validation

—— EARLY STOP ——

• Ending patience early   : 1000 – Patience for an early stop
• Min epochs for early stop   : 10 – Min. epochs before early stop
• Save best weights   : false – Keep best weights

—— ADAPTIVE LEARNING RATE ——

• LR decrease factor   : 0.5 – LR reduction factor
• LR increase factor   : 1.01 – LR Magnification Factor
• Error difference for LR decrease   : 1.01 – Error difference for LR reduction
• Minimum learning rate   : 0.00001 – Minimum learning rate
• Maximum learning rate   : 10000 – Maximum learning speed

—— NOTIFICATIONS ——

• Enable alerts   : true – Enable notifications
• Alert threshold   : 0.1 – Alert threshold
• Push notifications   : false – Push notifications
• Email alerts   : false – Email Notifications
• Sound alerts   : false – Sound notifications

——LOGING ——

• Training logs   : true – Training logs
• Log frequency reduction   : 10 – Logging frequency
• Example logs   : false – Example logs
• Result logs   : false – Result logs
• Load logs   : false – Loading logs
• Save logs   : false – Save logs

—— OTHER SETTINGS ——

• Universal output scaling   : true – Universal output scaling
• Fix indicator min/max   : true – Fix min/max of indicator window
• Max bars in window   : 5000 – Max bars in indicator window

II. Detailed description of parameters:

Prediction number (from 2(1 for TDif) to Out):
Predicting N-1 bars for T, TBin, and N bars for TDif. For most network types (T1, T2, T3, T4, T3Bin, T4Bin), the minimum value is 2. For difference types (T1Dif, T2Dif), the minimum value is 1.

Network filename prefix:
Prefix (initial part) of the file name that stores the weights of the trained neural network. Allows you to use different networks for different instruments or strategies while maintaining a clear name.

Filename postfix type:
The type of postfix (final part) of the file name. Determines what data will be automatically added to the file name to make it unique.

• NO: The postfix is not added.

• SYMBOL_TF: Adds a symbol and timeframe (eg: net_EURUSDH1).

• T_SYMBOL_TF: Adds network type, symbol and timeframe (eg: net_T1_EURUSDH1).

• T_ACT_SYMBOL_TF: Adds network type, activation type, symbol and timeframe (eg: net_T1_Tanh_EURUSDH1).

Use common folder for files:
Specifies the location of the files with the neural network weights. If the value true, files will be saved and loaded from the common folder of all terminals  Terminal\Common\Files. This is convenient for accessing the same network from different terminals. If  false – from a specific terminal folder  MQL5\Files .

Auto color by type:
Enable automatic selection of indicator drawing color depending on the type of neural network forecast. If the value  true, parameter Manual color ignored.

Manual color (if AutoColor false):
Sets a custom color for drawing the indicator. Active only if the parameter  Auto color by type  disabled (false).

—— NETWORK STRUCTURE —— 

Network type:
The type of neural network architecture and purpose. Each type is optimized for a specific market and analysis style. (Note: Read III. Technical description of the types of networks used).

• T1 (Normalized Analysis): Forex (majors), indices, metals.

• T1Dif (Difference Analysis): Cryptocurrencies, commodities, forex (minors).

• T2 (Context-Aware): Forex (crosses), metals, indices.

• T2Dif (Context-Aware Difference): Forex (crosses), metals, indices.

• T3 (Trend Detection): Forex (majors), commodities, indices.

• T3Bin (Binary Trend): All markets (training).

• T4 (Momentum Detection): Cryptocurrencies, Commodities, Metals.

• T4Bin (Binary Momentum): Cryptocurrencies, US stocks, commodities.

Activation preset:
A predefined set of activation functions and parameters for network layers. Allows you to quickly select a proven configuration.

• Auto: Automatic selection depending on the selected network type (Network type).

• Manual: Manually adjust the activation and scaling functions.

• // Basic & Recommended:

• Standard: Tanh-Tanh[-1,1] (T1,T2,T3,T2Dif) Gradient off

• Classic: Sigm-Sigm[0,1] (T3,T3Bin,T4Bin) Gradient off

• Difference: LRelu-Linear[-1,1] (T1Dif,T4,T2Dif) Gradient 0.1.

• BinaryMomentum: Relu-Sigm[-1,1]-[0,1](T4Bin,T3Bin) Grad 0.08

• // Advanced:

• Asymmetric: Tanh-Tanh[-1,1]-[0,1](T1,T2,T3) Grad off

• ReLUNetwork: Relu-Relu[-1,1](T4,T4Bin) Grad 0.1

• Regression: Tanh-Linear[-1,1](T1Dif,T2,T2Dif,T4) Grad off
  

• MixedAsymmetric: Tanh-Sigm[-1,1]-[0,1](T2,T3) Grad off
  

• // Alternative:

• Alternative: Tanh-Tanh[0,1](T1,T2,T3) Grad off

• ReLURegression: Relu-Linear[-1,1](T1Dif,T4,T2Dif) Grad 0.12

• LeakyReLU: LRelu-LRelu[-1,1](T4,T4Bin,T1Dif) Grad 0.1

• FullyLinear: Linear-Linear[-1,1](Exp,T1Dif,T4) Grad 0.15

• // Experimental:

• Hybrid: Sigm-Tanh[0,1]-[-1,1](T2,T3,T2Dif) Grad off

• ReLUSigmoid: Relu-Sigmoid[0,1](T4Bin,T3Bin) Grad 0.1

• ComboMomentum: Relu-Tanh[-1,1](T4,T1Dif) Grad 0.1

• Experimental: Sigm-Linear[0,1]-[-1,1](Any,untested) Grad off

• ComboLeaky: LRelu-Tanh[-1,1](T4,T1Dif,T2Dif) Grad 0.1

• ComboMixed: Tanh-Sigm[-1,1]-[0,1](T2,T3,T3Bin) Grad off

Loss function type:
The type of error function (loss function) that the neural network tries to minimize during the learning process.

• MSE (Mean Squared Error): Sensitive to outliers.

• MAE (Mean Absolute Error): Mean absolute error. Less sensitive to outliers.

• L_HUBER: HuberLoss. A compromise between MSE and MAE.

• BinCE (Binary Cross-Entropy): Binary cross-entropy. Designed for binary classification (output in the range [0,1]).

Input layer size:
The size of the input layer of the neural network. Determines the number of neurons that receive input data from the market.

Hidden layer 1 size:
The size of the first hidden layer of a neural network.

Hidden layer 2 size (0 = disabled):
Size of the second hidden layer. If set to 0, this layer is disabled and the network becomes a three-layer (input-hidden-output) network.

Output layer size (from 2, from 1-for TDif):
Output layer size. Determines the dimensionality of the neural network forecast. Usually from 2 and more. For difference types of networks (T1Dif, T2Dif) it can be from 1.

Network comment:
An arbitrary text comment that will be saved along with the network. Can be used to make notes about the training goal, data, etc.

—— MANUAL ACTIVATION SETTINGS —— 

Hidden layer activation:
Activation function for all hidden layers of the network. Active only when preset is selected    Manual .

• Sigm: Sigmoid. output (0,1).

• Relu: Rectified linear block. output max(0,x).

• LRelu: Leaky ReLU. Allows small negative values to pass through.

• Linear: Linear activation. output x.

• Tanh: Hyperbolic tangent. output (-1,1).

Output layer activation:
Activation function for the output layer of the network. Only active when a preset is selected.  Manual .

Input scaling:
A method for scaling input data before feeding it into the network.

Output scaling:
A method for scaling network output.

Gradient limiting:
Enable gradient clipping. Helps combat the “exploding gradients” problem during training.

Max gradient value:
Maximum absolute value of the gradient. Used if enabled  Gradient limiting .

—— LEARNING PARAMETERS ——

Start learning after loading:
If enabled ( true ), the training process will start immediately after the indicator is loaded onto the chart.

Use training date:
Whether to use a specific date for the training data slice. If off ( false ), training will end with the current date (minus the output layer).

Training date:
The date on which the training dataset ends. Any bars after this date will not be used for training.

Max training epochs:
Maximum number of training epochs. One epoch is a full pass through the entire training dataset.

Max training samples:
The maximum number of examples (samples) that will be used for training from the entire available history. In the T4, T4Bin strategies, only those that satisfy the conditions will be selected from this number of examples.

Bars between samples:
The minimum number of bars between two consecutive samples in a data set. Helps increase the diversity of the data.

Use adaptive learning rate:
Enable adaptive learning rate change. The algorithm will automatically increase or decrease the learning rate during training.

Learning rate:
Initial learning rate. Determines the “step” with which the neural network adjusts its weights during training.

Target error:
Target error value. If the error on the training sample reaches this value, training will stop.

—— VALIDATION SETTINGS —— 

Validation period (bars):
The validation period size in bars. Data for this period is not used in training and is used to check the quality of the network. If 0, validation is disabled.

Use validation criteria for early stopping (and saving best weights):
Whether to use a criterion on the validation set to stop training early and keep the best weights.

Validation mode:
The criterion by which the quality of a model on a validation sample is determined.

• Binar: Binary accuracy of the direction forecast.

• Profit: A hypothetical profit calculated as the difference between the bars in points of the current instrument (excluding the spread and commissions and other market conditions).

• NetError: Network error (MSE, MAE, etc.) on validation sample.

Validate only ‘Prediction number’ (for Binar,Profit):
If enabled, validation by criteria    Binary    And    Profit    will be held only for the bar specified in    Prediction number .

Validation prediction threshold:
Threshold value for forecast. Used in modes    Binary    And    Profit    to determine whether the prognosis is considered positive or negative.

—— EARLY STOPPING ——

Early stopping patience:
The number of epochs during which the error on the validation set may not improve before early stopping is triggered.

Min epochs before early stopping:
The minimum number of training epochs that must be completed before the early stopping mechanism kicks in.

Save best weights:
Whether to keep the weights of the best performing model rather than the last weights after training.

—— ADAPTIVE LEARNING RATE ——

LR decrease factor:
A multiplier to reduce the learning rate if the error has stopped decreasing significantly.

LR increase factor:
A multiplier to increase the learning rate if the error is steadily decreasing.

Error difference for LR decrease:
The threshold for error growth to initiate a decrease in learning rate.

Minimum learning rate:
The lower bound on the learning rate, below which it cannot be reduced.

Maximum learning rate:
An upper limit on the learning rate, beyond which it cannot be increased.

—— NOTIFICATIONS ——

Enable alerts:
Enable all notifications.

Alert threshold:
The forecast threshold value that, if exceeded, will generate an alert.

Push notifications:
Send push notifications to the MetaTrader mobile application.

Email alerts:
Send email alerts.

Sound alerts:
Play sound notification.

—— LOGGING ——

Training logs:
Enable logging of the learning process.

Log frequency reduction:
How many times to reduce the frequency of logging. For example, a value of 10 means that logging will occur every 10th epoch.

Example logs:
Keep detailed logs for each training example.

Result logs:
Keep logs of final results.

Load logs:
Keep logs of the network loading process from a file.

Save logs:
Keep logs of the network saving process to a file.

—— OTHER INDICATOR SETTINGS ——

Universal output scaling:
Apply universal scaling to output data for display in the indicator window. Alert threshold will be applied to the universal scale.

Fix indicator min/max:
Fix the minimum and maximum values of the indicator scale to improve visualization.

Max bars in indicator window:
The maximum number of history bars for which the indicator will calculate and display its values.

III. Technical description of the types of networks used: