Every indicator has its advantages and disadvantages. Trending ones show good signals during a trend, but lag during a flat. Flat ones thrive in the flat, but die off as soon as a trend comes. All this would not be a problem, if it was easy to predict when a flat changes to a trend and when a trend changes to a flat, but in practice it is an extremely serious task.

What if you develop such an algorithm, which could eliminate an indicator's flaws and enhance its strengths? What if such an algorithm could improve the operation of a trend indicator during a flat, but at the same time increase the efficiency during a trend, and fix the signals of a flat indicator during a trend and perfect them during a flat?

"Deep" is that very algorithm, which enhances the advantages of an indicator and minimizes its disadvantages. In short, this algorithm makes an indicator better. In an attempt to merge neural networks with a standard indicator, to my surprise, I received quite interesting results.

On the concept of "Deep".

The specified task, due to its non-reproducibility and complex regularity, is best solved by non-linear algorithms. Neural networks are a very versatile and powerful tool, which allows to achieve a lot, if the task is set correctly and the solution is formed reasonable. The algorithm is based on this very tool.

In fact, the algorithm measures the optimal periods of a moving average on all areas, depending on the state of the trend and the flat. On a flat area, the algorithm adapts the period of the moving average towards the flat, on a trend area - towards the trend. By increasing the depth of the "Deep", you can increase the flexibility of the adaptive properties of the algorithm.

The architecture of the network and the heuristic search algorithm were designed specifically for this task. Multilayered structure and the amount of neurons in a layer are regulated dynamically, you only need to specify the maximum number of neurons in a layer, the maximum size of a layer and the maximum number of layers.

Description of the settings

• Graphic objects prefix name - graphic objects prefix name (e.g., panel name).
• Timeframe - timeframe.

Moving average

• Period - period of the moving average.

The "Deep" algorithm

• Deep - needed to regulate the depth of the period of the moving average.
• Deep Smoothing - smoothing of the result.
• Deep Sensitivity - algorithm sensitivity.
• Spread in points - spread size for fine-tuning the heuristics.

Neural network

• Max: Number of Sensors - maximum number of sensory neurons.
• Max: Layer Size - maximum number of neurons in a layer.
• Max: Number of Layers - maximum number of layers.
• Max: ratio of signal deformation - ratio is used for deformation signals.
• Max: Signal range - signal changing range.
• Signal digits - accuracy of the signals.

Calculations

• Power - power of the calculations, set in percentage from 1 to 100.
• Number of Bars - number of bars for calculations.

Signals

• Signal mode - signals mode.
• Q Method - adaptivity calculation method. These algorithms are used to improve the adaptive characteristics of signals.

Visual

• Show Mode - indications display mode.
• Up Arrows Code - up arrows code.
• Dn Arrows Code - down arrows code.
• Arrows Vertical Shift (in points) - arrows offset on vertically.

Notifications

• alerts? - displays a message in a separate window?
• Play Sound? - play sound file?
• File Name To Sound Play - sound file name.
• Send To Mail? - send to email?
• Mail Header - mail header.
• Notifications? - sends push notifications to the mobile terminals?

The panel

• Panel? - display the panel?
• Starting X - starting position horizontally (in pixels).
• Starting Y - starting position vertically (in pixels).
• Size multiplier - panel size multiplier.
• Color - panel color.
• Transparent % - transparency of the panel (from 10 to 100).
• Border Color - panel border color.
• Slider color - slider color.
• Txt color - base text color.
• Txt 2 color - other text color.