Replies: 2 comments 4 replies
-
Hei Jacob! I think I get your point, but let me rewrite it here in my own words just in case: ALPypeRL will call If you do so, I think you are right. I have thought about this before. In the simulations/models I have build, the action was required at For now, the only recommendation I can give you to solve your issue is to add an if-statement in your
if (time() == 0) {
// Do nothing
} else {
// The rest of your code here
// [...]
}
if (time() == 0) {
// Just return an empty observation
return <an-empty-observation>;
}
return <the-observation>;
if (time() == 0) {
// No reward
return 0;
}
return <the-reward>; The downside of this approach is that the neural network will associate a reward of 0 for any random action at the beginning. Maybe another good recommendation is to add As I said, I'll consider adding this feature in future releases. I'll keep you up to date. Happy coding! |
Beta Was this translation helpful? Give feedback.
-
Hei @jacobtw2! Please check the latest alpyperl-1.0.0 version. I think I managed to fix this, but I need your input to validate it. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
An issue I have noticed when trying to work with ALpypeRL is that the model needs to be in a position to perform the first action immediately upon start up.
I have a model where I request an action each time a new order reaches a certain point in the flowchart logic and the action decides what to do with the order. Calling upon reset would cause an issue because there is no order that needs it yet
Is there a way we can tell the simulation to run until the first action before returning back to the python side when reset is called. i have tried calling runFast on startup but this did not work.
Unfortunately, it is not even possible to manually set the model to a state where there is an order to process because any item for inserting an agent into a flow chart block such as inject or take is an event. Even though it happens in 0 time, it still requires time to move forward which it won't do until after going to the python side
Does anyone have any ideas on how to resolve this issue?
Beta Was this translation helpful? Give feedback.
All reactions