Deep Moving Average MT4

Fixed bugs in the Neural Network.

Model: C117

1. The indicator has been completely redesigned.

2. New neural network architecture (NN).
2.2. Convolutional layers have been added.
2.3. Improved architecture of hidden layers.
2.5. Now the structure of the NN is set by fixed values.
2.6. Added PreSets of the NN structure.
It is configured using the input parameter 'PreSet'.
If 'PreSet' = 'CUSTOM SET', then the structure is configured manually using a group of input parameters "Custom Set".

3. The MTF mode has been removed.
This is done to improve the performance and stability of calculations.

4. The learning system has been improved.
4.1. An adaptive learning system has been implemented.
4.1. The learning process has been optimized.
4.2. Now the learning process is saved when the indicator is restarted. This allows you to continue training the indicator after it is restarted.
4.3. New parameters for learning:
- 'Signal Mode' - signal calculation method (for line color, arrows and learning).
Options:
'Direction (MA line)' - the line direction is used.
'Crossing (Price with MA line)' - the intersection of the bar closing price with the MA is used.

- 'Criterion(for Quality)' - criteria for learning (adapting process).
Options:
'Balance max' - the learning system will strive to optimize net profit.
'Profit Factor max' - the learning system will strive to optimize the profit factor.
'Recovery Factor max' - the learning system will strive to optimize the recovery factor.

5. The "Deep" algorithm has been improved.
5.1. Unnecessary parameters have been removed.
5.2. New parameter 'Sensitivity in % [1...100]' - the sensitivity of the MA period change. More precisely, it is the rate of change.
The lower the value, the slower the MA period will change. Use to avoid fast changes of the MA line.
5.3. New parameter 'Period(in bars) for Re-Adapt(0-Adapt once)' - time period (number of bars) for automatic start of repeat learning (adaptation).
5.4. New parameter 'Maximum Iterations [1...)' is the maximum number of iterations for the running learning process (adaptation).
5.5. Now the MA period is set through the minimum and maximum values.
For reference: if 'Minimum Period [1...)' = 'Maximum Period [1...)', then the "Deep" algorithm will not be activated and the training system (adaptations) will not be used.

6. Improved visual panel.
6.1. Visual improvements.
6.2. Increased functionality.
6.3. Added more visual information.
6.4. A button has been added for manually stopping the learning process (adaptation), as well as for starting the learning process (adaptation) again.

- Improved: the 'Visual Info' function.
Added new input parameter 'Starting Bar for Arrows'.
If value = 'Current Bar', then the arrows are drawn from the current bar.
If value = 'Closed Bar', then the arrows are drawn from the last closed bar (use this to wait for more stable arrows).

- Improved: the 'Notify' function.
-- The notification text was shortened and improved.
-- Added new input parameters:
1) 'Use Prefix Info'.
If = true, then the value of the input parameter 'Graphic objects prefix name' will be added to the notification description.
2) 'Use Currency Pair Info'.
If = true, then the chart Symbol (Currency Pair) name will be added to the notification description.
3) 'Use Chart Period Info'.
If = true, then the chart Time Frame will be added to the notification description.