Skip to content

DynamicsLoss does not use z_pred_loss while training #4

Description

@famishedrover

When compute_dynamics_loss is called in world_model.py, it is fed in an argument called preds returned from dyn_model.predict(). For z prediction, preds have keys z_dist and z_hat_probs and there is no key z.

When computing the dynamics loss, specifically the loss for z, the code checks for whether z is in the list of keys preds which it never is (because the keys are z_dist and z_hat_probs). For this reason the z_pred_loss is never computed & used for training.

A fix should be made here, something like :

if 'z_dist' in preds: ...

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions