General conditions.
The Expert Advisor can be trained to work on ANY instrument.
DO NOT TRY TO TEST NEURAL NETWORKS WITHOUT TRAINING!
it is enough that the balance chart after training is horizontal.
Generating a training base is extremely simple.
ready-made training is available for USDCHF.
For proper operation, it is necessary that there is only one account on one account! adviser and only on one! the instrument. Otherwise, the balance control will not work.
The initial deposit is from 200 ue.
Parameters:
– OrderComm: in the order, it will be supplemented with a sign of the mode of operation and depth of analysis.
– TypeOfWork: Textbook Marketplace.
“Study” – in this mode, it is necessary to achieve a schedule with at least a small growth. Without pits. This mode can also be used for work, but there will be only 1 order in the market at a time.
“Work” – The database is loaded periodically. In case of simultaneous training in the tester and constant pumping of higher-quality training.
“Market” – only for Market acceptance. It is necessary to switch.
[spoiler title=”Read More…”]
– MYpercent: …
– SL: If set to 0 in Training mode, the value from 40 to 110 will be automatically assigned and will give too many trades, slow training and distortion of images. With MaxOrders>1 and the operation mode is set to 5000 forcibly.
– StartTrail: Not used in Training mode. Optimal =0 (automatically).
– Trail:…
= Neuro param
– Threshold 1: threshold for noise elimination on 1 hidden layer
– ThresholdOut: threshold for screening out unsuitable images. During training, it is forcibly set to 0.
– Trade interval: every transaction … For training, H is recommended1. Then in Operation it is forcibly set to 1 minute.
– Depth Analyzes: the interval of bars for analyzing the price picture. The last 50 bars from the selected TF are loaded. Example: with TF M 6 snapshot => 50*6/60= 5 hours. A drawing of such length is processed for memorization. Too large a TF will result in non-repeating snapshots. Too small – it will give a lot of very similar pictures. The database is formed by this parameter. If you select a different value, the database will be created and full training is required. The databases for each TF are not deleted and once created, you can continue to train and use. If you want to train the network again, you need to delete these files. The location of the files will be indicated in the log when the Expert Advisor is launched. Something like this – c:Users[YOUR_USERNAME]AppDataRoamingMetaQuotesTerminalCommonFilesNeuro. The attached databases should be located there. Example: data_w1_6_USDCHF.csv, data_w2_6_USDCHF.csv. These files contain “_6_” in the name and will be connected when “Depth Analyzes“=M 6.
= Learn mode
– LearnEpoch: parameter for setting the number of repeated optimization runs in Training mode.
– speed: with initial training, several training cycles (5-10) can be done at a speed of 0.5-0.9. The following at 0.2-0.5. Etc. When an entry of the form w 1[XXX][YYY]N.NNNNNN=ZZ.ZZZZZZZZ appears in the logs and stops improving the training of some area, you need to set the speed to 0.1 or less.
= Dual mode
– MaxOrders: If the value is >1, the insurance strategy will be launched. After all, the neural network does not guarantee the repetition of the trained outcome after a similar snapshot, but assumes the greatest probability. In the Training mode, it is forcibly set to =1 because training takes place on the data of only 1 simultaneously running order.
– Multiply: volume increment factor. Optimally 2.
– Distance: the distance between orders of the same direction. Increases with each new order of the same direction.
– StepProf: another additional system, but no less important. Closing all orders when the balance gain is reached on… The target value of the trigger is indicated in the comment on the graph as “Next“
= Other
– Pass: a key to remove restrictions. It is not used for demo accounts.
The price is negotiable.
[/spoiler]