Discussing the article: "Neural networks made easy (Part 39): Go-Explore, a different approach to exploration" - page 2

New comment
 
Dmitriy Gizlyk #:


Hello Dimitri.

Can this approach be identified with "memory optimisation algorithm".

Where the algorithm remembers its search history or is there some other subtle point that I didn't catch?

 
Why are there three options to buy/sell/close all positions in the agent's random actions block? Why is there no option "hold a profitable position (do nothing)". The random action is performed at the opening of an hour candle, if the deal is in the plus, why do you have to close it? You can sometimes get 300 pips in an hour.
 
npats2007 #:
Why are there three options to buy/sell/close all positions in the agent's random actions block? Why is there no option "hold a profitable position (do nothing)". The random action is performed at the opening of an hour candle, if the deal is in the plus, why do you have to close it? You can sometimes get 300 pips in an hour.

Note that we semplicate their 4 actions. An agent can perform one of 3 actions:

0 - buy
1 - sell
2 - close all open positions
and 3 - wait, do not perform any actions.

Action 3 refers to waiting for a suitable state of opening or closing actions. Including the situation of profit accumulation described by you. Only in this case the Expert Advisor does not perform trading operations. Therefore, the processing of such an action is deliberately omitted.

 
Dmitriy Gizlyk #:

Note that we sample their 4 actions. An agent can perform one of the 3 actions:

0 - buy
1 - sell
2 - close all open positions
and 3 - wait, take no action.

Action 3 refers to waiting for a suitable state of opening or closing actions. Including the situation of profit accumulation described by you. Only in this case the Expert Advisor does not perform trading operations. Therefore, the processing of such an action is deliberately omitted.

Got it. Thank you.

 

Still opens only buy and fills on every bar. Can you tell me how to fight this problem? Does anyone have it perform any meaningful actions?

Files:
6u3_m6ydd.png  62 kb
 
I think I understand him. He thinks that the only profitable strategy is to open and wait without looking at drawdowns. Since the euro is on the upswing, he's going short. It will be fun if he starts opening short when the trend changes! If I live to see it, I will definitely check it out!
 

How much phase 2 training did you give him? How many times did you run phase 2?

And what was the error when you went to phase 3?

And I'm getting a higher error every time I run phase 2. Is that normal? I understand that in a neural network the error should decrease during training, but here it turns out the opposite....

 

I got this error.

2023.05.07 20:04:44.281 Core 01 pass 359 tested with error "critical runtime error 502 in OnTester function(array out of range, module Experts\GoExploer\Faza1.ex5, file Faza1.mq5, line 223, col 12)" in 0:00:00.202

//--- copy cell

actions[action_count] = act;

Base[action_count].total_actions = action_count+StartCell.total_actions;


how to solve it?


 
Viktor Kudriavtsev neural network the error should decrease during the learning process, but here it turns out to be the opposite....

If the error is constantly growing, try to reduce the training coefficient.

 
Zhongquan Jiang #:

I got this error.

2023.05.07 20:04:44.281 Core 01 pass 359 tested with error "critical runtime error 502 in OnTester function (array out of range, module Experts\GoExploer\Faza1.ex5, file Faza1.mq5, line 223, col 12)" in 0:00:00.202

//--- copy cell

actions[action_count] = act;

Base[action_count].total_actions = action_count+StartCell.total_actions;


how to solve it?


What's period of study?

123
New comment