Examples: Using Neural Networks In MetaTrader - page 11

[Farhan Trader]

Has any one updated and using this EA in auto optimization,.... because the version is too old and many function is not working ...?

[[Deleted]]

In ann_load

ann = f2M_create_from_file_string (path);

and in ann_save

ret = f2M_save_string (ann, path);

changed functions to string functions. Now works for me. Attached is my updated example code for this great work

[MataNui]

Simply isnt working.

Install went fine. No errors logged. Training files are saved correctly. It just doesnt do anything. The results are identical withthe filter on and off.

[Aldo Marco Ronchese]

Hi Guys i wonder if you could help . Ive been hacking this a bit but a few issues perhaps already covered . I have to edit the file to set the learning rate as i couldnt set it through the included dll converter . Also limited to 2 layers . Cant think of a way without making file in different system like python with 3 or more layers and then using file as it ignores settings and uses files specs? So what i need help with is setting learning rate and number of layers above 2 .. perhaps can be hard coded a new command to create ann for 3 .. the extra layer makes it deep learning and this could make a huge difference. Thanks for your time

[Aldo Marco Ronchese]

Stanislav Korotky #:

According to some researchers (for example, read here) 2 hidden layers are sufficient for classical MLP in almost all tasks. Increasing number of layers will only slow down performance without result improvement.

This is opinion . The way it works a lot of neurons are waisted . The randomization is actually where the intelligence ls . Some initial randomizations are closer to being optimized than others . Others are so far out they will never train . Having more layer gives more focus and and context to the data . This helps it to learn faster . To make an analogy each layer is like an attribute. Each layer is a filter . 2 dimensions vs 3 dimensions.. you can represent 3 in 2 but its easier to just have 3  . All this to justify request for 3 layers pls . It is a big learning curve for me to change the source and recompile. After all the fabulous work you have done this plus the ability to change the learning rate and possibly the momentum would complete the package . So far you ea converted to my ideas trains to 99.9% accurate on any data . The new data however it falls short . I feel the extra layer will provide context  . I will also give you the EA if you could help me . It is your work plus fann plus 2 years of my own research  

[Aldo Marco Ronchese]

This is what i was able to do a year ago .. since then no progress and i think this is ley , extra layer . Thank you so much for your excellent work !