xpag.agents.flax_agents.td3.td3.FlaxTD3

class FlaxTD3(observation_dim, action_dim, params=None)

Bases: Agent

Jax implementation of TD3 (https://arxiv.org/abs/1802.09477), without delayed policy updates. This version assumes that the actions are between -1 and 1 (for all dimensions).

Methods

load
save
select_action
train_on_batch
value
write_config