About executing commands: Is Python train.py training action recognition or only training human skeletal points

[allert11]

Select the configuration file as joint.py, and after training, generate the best weights for testing the dataset. The output pkl file does not have labels, which means there are no output categories during testing. So, did we not perform action classification during the training process? It's just a simple exercise to determine whether the bone points are accurate