# @package _global_
img_size: 64

# Dreamer defaults
rssm: {ensemble: 1, hidden: 1024, deter: 1024, stoch: 32, discrete: 32,  norm: layer, std_act: softplus, min_std: 0.1,  single_obs_posterior: true, } # act: elu,
discount_head: {layers: 4, units: 512, norm: none, dist: binary} # act: elu
reward_head: {layers: 4, units: 1024, norm: layer, dist: twohot} # act: elu
kl: { free: 1.0, forward: False, balance: 0.85, free_avg: False, }
loss_scales: {kl: 0.6, reward: 1.0, discount: 1.0, proprio: 1.0}
model_opt: {opt: adam, lr: 1e-4, eps: 1e-8, clip: 1000, wd: 1e-6}
replay: {capacity: 20e6, ongoing: False, minlen: 48, maxlen: 48, prioritize_ends: False}
decoder_inputs: stoch
image_dist: mse
# Actor Critic
actor: {layers: 4, units: 1024, norm: layer, dist: normal, min_std: 0.1 } # act: elu 
critic: {layers: 4, units: 1024, norm: layer, dist: twohot } # act: elu, 
actor_opt: {opt: adam, lr: 3e-5, eps: 1e-5, clip: 100, wd: 1e-6}
critic_opt: {opt: adam, lr: 3e-5, eps: 1e-5, clip: 100, wd: 1e-6}
discount: 0.99
discount_lambda: 0.95
slow_target: True
slow_target_update: 100
slow_target_fraction: 1
slow_baseline: True
reward_ema: True

acting_reward_fn: env_reward
clip_rewards: identity

batch_size: 48 
batch_length: 48 
imag_horizon: 16
eval_state_mean: False

precision: 16
train_every_actions: 10
only_random_actions: False