I made a new Grid martingale Experiment.
Only 1 input parameters: Desidered monthly profit

It's not a joke, I think that machine learning could bring impressive results, but it takes a great amount of time to compute the neural network. I'm already using the mt5 cloud, but it takes hundreds of dollars for 1 year optimization...
In the following example I trained the net over real tick data in 2016 and test it in forward test over 2017.
In few words, I gave to the algo all datas from 2016, and then I used this data to backtest 2017. I don't know if I will go on developing further more, it's too expensive for me.
I'm attaching the compiled version of the EA, it works only for 2017, it's a non sense to try in other periods.
The best things would be to train 3 or 6 month to predict the next month, but...too much work (and money) for me at the moment. I tried also to use Microsoft Azure ML platform, but it's free only for some training works, then it become a bit expensive.

Feel free to download the attached EA (already trained in 2016 to trade in 2017).

Backtest is on EURUSD, real tick data, H1, Metaquotes server

Attached Files