regression.yaml

model: ${leo_gn}
enet: ${enet_rff}
n_update: 5000
n_inner_update: 1
seed: 0
n_hidden: 128
n_samples: 100
clip_norm: True
lr: 1e-4

n_disp_step: 200
show_plot: True
save_plot: False
save_model: False

enet_basic:
  class: regression.EnergyNetBasic
  tag: enet_basic
  params:
    n_hidden: 128

enet_rff:
  class: regression.EnergyNetRFF
  tag: enet_rff
  params:
    n_hidden: 128
    sigma: 1.
    encoded_size: 128

unroll_gd:
  class: regression.UnrollGD
  tag: unroll_gd
  params:
    n_inner_iter: 10
    inner_lr: 1e-3
    init_scheme: zero # zero, gt

unroll_gn:
  class: regression.UnrollGN
  tag: unroll_gn
  params:
    n_inner_iter: 10
    inner_lr: 1 # standard GN is 1.0
    init_scheme: zero # zero, gt

leo_gn:
  class: regression.LEOGN
  tag: leo_gn
  params:
    n_sample: 100
    temp: 1e9
    min_cov: 1e-3
    max_cov: 10.0
    n_inner_iter: 10
    init_scheme: zero # zero, gt

dcem:
  class: regression.UnrollCEM
  tag: dcem
  params:
    n_sample: 100
    n_elite: 10
    n_iter: 10
    init_sigma: 1.
    temp: 1.
    normalize: True

leo_cem:
  class: regression.LEOCEM
  tag: leo_cem
  params:
    n_sample: 100
    temp: 1.
    min_cov: 1e-3
    max_cov: 10.0
    cem_n_sample: 100
    cem_n_elite: 10
    cem_n_iter: 10
    cem_init_sigma: 7.
    cem_temp: 1.
    cem_normalize: True