Graph execution error: indices does not index into param shape

Hi, I am a beginner in using tensoflow and i built a classification model relying heavily on tf_agent.
Some time ago i increased the number of used datasets and since then i get the following error message.

Error Message:
“W tensorflow/core/framework/] OP_REQUIRES failed at : INVALID_ARGUMENT: indices[4] = [4, 200, 66] does not index into param shape [10,200,200], node name: GatherNd”

This error always occurs after step 8 when running the code. (see github link)
Changing the number of training loops and number of datasets changes also the step count till the error occurs. Couldn’t to find a direct correlation between those two yet as well as the cause why it occurs in the first place.

Error reproducable: yes
Used python: 3.11.2
Used code: KI-Error/ at main · MarvinGe/KI-Error · GitHub
Used packages: KI-Error/requirements.txt at main · MarvinGe/KI-Error · GitHub
Detailled error description: KI-Error/Error_Description.png at main · MarvinGe/KI-Error · GitHub

Any kind of help appreciated.
Thank you

Topic can be closed.
Problem solved. Reward distribution needs to have same number as min/max range of labels.

1 Like