dreamer_discrete_world_model

→DreamerForWorldModeling

dreamer_discrete_world_model(pretrained: bool = False, overrides: object = {})

source edit

Construct the discrete-control configuration with all three objectives.

Model Size

Parameters

pretrainedbool= False

No weights are published for this family; passing True raises.

**overridesobject= {}

Optional DreamerConfig field overrides.

Returns

DreamerForWorldModeling

The discrete-control trunk plus the objectives.

Notes

Reference: Hafner, Lillicrap, Ba, and Norouzi, "Dream to Control: Learning Behaviors by Latent Imagination", ICLR, 2020 (arXiv:1912.01603), Appendix A.

pcont=True here, so forward needs its discounts argument — the discount head is "trained towards the soft labels of 0 and gamma", which is what this family already does with that tensor.

Examples

>>> from lucid.models.generative.dreamer import dreamer_discrete_world_model
>>> model = dreamer_discrete_world_model(action_dim=18).eval()
>>> model.config.pcont, model.config.pcont_scale
(True, 10.0)

Used by 1

lucid.models.generative.dreamer