Don't tell me then that TA doesn't work - page 13

Posted an EA last night, didn't comment right away. I've never written NS, so my amateurish bicycle:

1. Weights are first normalized so that |X[1]| + |X[2]| + |X[3]| + |X[4]| = 1. During optimization X[i] can take all values from the interval [-1, +1].
2. Parameter Smoke <= 1.
3. In the perceptron the price data is given as relative returns: D[i] = Open[(i - 1) * shift] / Open[i * shift] - 1.
4. Before calculation (implicitly by code), the price data(D[i]) is normalized so that |ResX| = |X[1] * D[1] + X[2] * D[2] + X[3] * D[3] + X[4] * D[4]|= 1, and there is always a set of weights {Y[1], Y[2], Y[3], Y[4]}, |Y[1]| + |Y[2]| + |Y[3]| + |Y[4]| = 1, for which |ResY| = |Y[1] * D[1] + Y[2] * D[2] + Y[3] * D[3] + Y[4] * D[4]| = 1 (maximum possible).
5. The signal strength Res = X[1] * D[1] + X[2] * D[2] + X[3] * D[3] + X[4] * D[4] is compared to Smoke: |Res| < Smoke - CLOSE, Res >= Smoke - BUY, Res <= -Smoke - SELL.

With this scheme there is a possibility to go from (+1, -1, 0) to MM, because Res is an implicit position volume. I will make such variant later.

Expert Advisor according to the MM scheme.

This is the rare "correct" case when the signal contains not only the direction of the position, but also its volume. For example, if the signal is weak, the volume is small, strong - large. In this case at each point of time the net position is "floating".

Here https://www.mql5.com/ru/forum/131739 explains it all - Reshetov's first post.

In the same thread Reshetov simply offered a program to automate the process described here https://www.mql5.com/ru/forum/131739.

Apparently people were too lazy to understand it here https://www.mql5.com/ru/forum/131739, so the man tried to clearly demonstrate

The real efficiency of this process https://www.mql5.com/ru/forum/131739.

And I have not inserted or corrected anything anywhere !