I had made this mistake in a model, masking array was longer than actions. The experiments trained fine but when running with policy I get this error about observations:
Array length not equal to model input size
I think it cuts the masking array to fit the size of actions.
Would be great if we throw an error in AL so users would know there is something wrong when running in random actions.
Also, if masking array was a jagged array instead one long array, we could do the check for each decision separately.
So for example here we can check that the first element of masking array is of size 4 and the second one is of size 3.
I had made this mistake in a model, masking array was longer than actions. The experiments trained fine but when running with policy I get this error about observations:
Array length not equal to model input sizeI think it cuts the masking array to fit the size of actions.
Would be great if we throw an error in AL so users would know there is something wrong when running in random actions.
Also, if masking array was a jagged array instead one long array, we could do the check for each decision separately.
So for example here we can check that the first element of masking array is of size 4 and the second one is of size 3.