×
type: MemoryGym
name: Endless-MortarMayhem-v0
frame_skip: 1
last_action_to_obs: False
last_reward_to_obs: False
obs_stacks: 1
grayscale: False
resize_vis_obs: [84, 84]
positional_encoding: False
reset_params:
start-seed: 200002
num-seeds: 1
max_steps: 2560
agent_scale: 0.25
agent_speed: 3.0
allowed_commands: 9
initial_command_count: 1
command_show_duration: [3]
command_show_delay: [1]
explosion_duration: [6]
explosion_delay: [18]
visual_feedback: True
reward_command_failure: 0.0
reward_command_success: 0.1
seed: 200002
reward_normalization: 0
positional_endocing: False
×
load_model: False
model_path: ./checkpoints/mg_17r/20220112-065616_2/mg_17r-208.pt
checkpoint_interval: 500
activation: relu
vis_encoder: cnn
vec_encoder: linear
num_vec_encoder_units: 128
hidden_layer: default
num_hidden_layers: 1
num_hidden_units: 384
vec_encoder_layer: linear
transformer:
num_blocks: 3
embed_dim: 384
num_heads: 4
memory_length: 256
positional_encoding: absolute
add_positional_encoding_to_query: True
layer_norm: pre
init_weights: xavier
gtrxl: False
gtrxl_bias: 0.0
gtrxl_swap: False
share_heads: True
max_episode_steps: 2561
ground_truth_estimator:
detach_gradient: False
×
n_workers: 32
worker_steps: 512
×
algorithm: PPO
resume_at: 0
gamma: 0.995
lamda: 0.95
updates: 50000
epochs: 3
n_mini_batches: 8
advantage_normalization: no
value_coefficient: 0.5
max_grad_norm: 0.25
learning_rate_schedule:
initial: 0.000275
final: 0.0001
power: 1.0
max_decay_steps: 10000
beta_schedule:
initial: 0.0001
final: 1e-06
power: 1.0
max_decay_steps: 10000
clip_range_schedule:
initial: 0.1
final: 0.1
power: 1.0
max_decay_steps: 10000
obs_reconstruction_schedule: {'initial': 0.0}
ground_truth_estimator_schedule:
initial: 0.1
final: 0.1
power: 1.0
max_decay_steps: 20000
share_parameters: True