tester got "xx pases not processed and returned to task queue" warning

[herrcrowley]

I am trying to run walk forward optimization usually my total steps list is around 220k results and I found this warning but test shows as completed, is this fine or should I think some results are going missing??

[Alexandre Sousa Bezerra]

herrcrowley: I am trying to run walk forward optimization usually my total steps list is around 220k results and I found this warning but test shows as completed, is this fine or should I think some results are going missing??

I have this same issue!!! But my forward test stops at 65.536 scenario. Doesn't matter how long I wait, tester doesn't go further.

[Fernando Carreiro]

@Alexandre Sousa Bezerra #: I have this same issue!!! But my forward test stops at 65.536 scenario. Doesn't matter how long I wait, tester doesn't go further.

It is a bug that has been identified and reported ... https://www.mql5.com/en/forum/457425#comment_50704805

Forum on trading, automated trading systems and testing trading strategies

New MetaTrader 5 Platform build 4040: Improvements and fixes

Fernando Carreiro, 2023.11.23 16:36

I would like to hereby confirm a bug in the Strategy Tester's forwarding testing, as discovered and described by the user tanner gilliland in the following 4 topics ...

• Can anyone see if they can reproduce a bug on their computer? It wont take long.
• Is there a way to run or rerun forward tests with the same inputs as a previous backtest?
• Why would the number of passes in the strategy tester determine whether a bug happens or not.
• How can I see what exactly went wrong when a forward optimization pass fails?

I personally run the test offered by the user to reproduce the issue and confirmed his findings.

My complete testing procedure and results, including log files can be found on the following consecutive posts:

• https://www.mql5.com/en/forum/457425#comment_50704805
• https://www.mql5.com/en/forum/457425#comment_50704819
• https://www.mql5.com/en/forum/457425#comment_50704852
• https://www.mql5.com/en/forum/457425#comment_50704895
• https://www.mql5.com/en/forum/457425#comment_50704950

Please note, that credit for discovering this bug goes to the user tanner gilliland!

[Alexandre Sousa Bezerra]

Fernando Carreiro #:

It is a bug that has been identified and reported ... https://www.mql5.com/en/forum/457425#comment_50704805

Thanks, Fernando!

I've read through your post, and I totally agree. Combining too many variables in a regular EA situation is just plain crazy.

My EA relies on a Deep Neural Network with a 6-7-7-3 layer/neuron setup, which means dealing with a whopping 129 weights. That's a real headache!

I get it when you say that Metatrader might not fix this unusual issue. I've already contacted the Service Desk, so let's cross fingers.

Thanks again!

Alexandre

[tanner gilliland]

I'm glad I'm not the only one who is trying to get this fixed! Right now there is no point in even forward testing for me because all mine always get stuck at 65536. If you do find yourself doing another forward test could you watch the number of passes that are shown being completed at the bottom? For me they get stuck temporarily at 512 and 1024. I don't know if it matters or if that only happens to me, but maybe its another clue to help the devs find and fix the problem.

[Alexandre Sousa Bezerra]

I'm glad too! Mainliy cause I know that this is not a code problem...

Take a look at my situation:

 

And see how my Cores look like

 

[Alexandre Sousa Bezerra]

What I do to avoid this problem is using higher step values.

For example: I have 129 weights to combine in my DNN. the Weighst must vary between -1 and 1. I first consider step = 0.1, what gives me a total scenarios normally around 40.000,

Then, I analyze the range of the best results (top 10 for example, or top 1000) in my forward test for each variable and refine the it for the next Tester. For example:

Instead of using a range [-1,1] for all  weights, I consider a range w0 = [-0.7,0.3], w1 =  [0.1,0.4] until w129... it reduces the total number of scenarios and allows me to consider a more precise step, for example, step=0.01, but always keeping in mind that I cant have too much scenarios.

I have my excel that automatically generates the step file after inputing thounsand scenarios. It helps a lot!

I

[tanner gilliland]

Alexandre Sousa Bezerra #:

What I do to avoid this problem is using higher step values.

For example: I have 129 weights to combine in my DNN. the Weighst must vary between -1 and 1. I first consider step = 0.1, what gives me a total scenarios normally around 40.000,

Then, I analyze the range of the best results (top 10 for example, or top 1000) in my forward test for each variable and refine the it for the next Tester. For example:

Instead of using a range [-1,1] for all  weights, I consider a range w0 = [-0.7,0.3], w1 =  [0.1,0.4] until w129... it reduces the total number of scenarios and allows me to consider a more precise step, for example, step=0.01, but always keeping in mind that I cant have too much scenarios.

I have my excel that automatically generates the step file after inputing thounsand scenarios. It helps a lot!

I

Try the latest beta version. I think they fixed it! It no longer gets stuck for me.

[Alexandre Sousa Bezerra]

tanner gilliland #:

Try the latest beta version. I think they fixed it! It no longer gets stuck for me.

It is true! Already taking advantage of it. Neural Network now faster finds best weights... just need to take care to avoid overfittings.

Thanks for you tip Tanner!