Raylab 0.8.6

New Features

• (modules): add new DDPG and MBDDPG
• (modules): add new SAC and MBSAC
• (modules): add stochastic models with parallelized forward pass for ensembles
• (modules): add stochastic policies and actor
• (agents): use new DDPG and SAC modules as defaults
• (agents): use new model-based DDPG and SAC modules as defaults
• (policy): check model, action dist, and exploration compatibility
• (policy): check module compatibility with action dist
• (policy): add model property
• (policy): add stochastic and deterministic action dist wrappers
• (pytorch): add reverse mode to TanhSquash
• (pytorch): add no-op initializer

Refactorings

• rename raylab.losses to raylab.policy.losses
• rename raylab.modules to raylab.policy.modules
• (policy): remove TargetNetworksMixin
• (tests): match test and package directory structures

Others

• (examples): use MBSAC in MBPO
• (examples): use MBSAC in MAPO
• (modules): rename parameter_noise option to separate_behavior
• (modules): move old NNs to v0 submodule
• (policy): allow subclasses to set dist_class before calling init