adding naive jax and libraries and make adjustments

engine/sweeps/model_mix.yaml

@@ -0,0 +1,84 @@
+method: random
+metric:
+  name: sweep/score
+  goal: maximize
+command:
+  - ${env}
+  - python
+  - -m
+  - engine.train
+parameters:
+  algo:
+    values: [ppo, a2c, dqn, qtable]
+  total_timesteps:
+    values: [30000, 50000, 80000]
+  seed:
+    values: [13, 42, 77]
+  n_products:
+    values: [8, 10, 12]
+  alpha:
+    distribution: uniform
+    min: 0.1
+    max: 0.6
+  lambda_coi:
+    distribution: uniform
+    min: 0.05
+    max: 0.6
+  robust_radius:
+    distribution: uniform
+    min: 0.0
+    max: 0.3
+  robust_points:
+    values: [3, 5, 7]
+  info_value:
+    distribution: uniform
+    min: 0.5
+    max: 2.0
+  revenue_weight:
+    values: [0.005, 0.01, 0.02]
+  learning_rate:
+    distribution: log_uniform_values
+    min: 1.0e-5
+    max: 1.0e-3
+  gamma:
+    values: [0.97, 0.99, 0.995]
+  buffer_size:
+    values: [20000, 50000, 100000]
+  batch_size:
+    values: [128, 256, 512]
+  tau:
+    values: [0.002, 0.005, 0.01]
+  train_freq:
+    values: [1, 4, 8]
+  learning_starts:
+    values: [500, 1000, 3000]
+  n_steps:
+    values: [512, 1024, 2048]
+  n_epochs:
+    values: [5, 10, 20]
+  gae_lambda:
+    values: [0.9, 0.95, 0.98]
+  clip_range:
+    values: [0.1, 0.2, 0.3]
+  ent_coef:
+    values: [0.0, 0.005, 0.01]
+  target_update_interval:
+    values: [500, 1000, 2000]
+  exploration_fraction:
+    values: [0.1, 0.2, 0.3]
+  exploration_final_eps:
+    values: [0.01, 0.03, 0.05]
+  action_levels:
+    values: [7, 9, 11]
+  action_scale_low:
+    values: [0.75, 0.8, 0.85]
+  action_scale_high:
+    values: [1.15, 1.2, 1.25]
+  q_lr:
+    values: [0.03, 0.05, 0.1, 0.2]
+  eps_start:
+    value: 1.0
+  eps_end:
+    values: [0.02, 0.05, 0.1]
+  eps_decay:
+    values: [0.999, 0.9995, 0.9999]

engine/sweeps/models_only.yaml

@@ -0,0 +1,85 @@
+method: grid
+metric:
+  name: sweep/score
+  goal: maximize
+run_cap: 4
+command:
+  - ${env}
+  - python
+  - -m
+  - engine.train
+parameters:
+  algo:
+    values: [ppo, a2c, dqn, qtable]
+  seed:
+    value: 42
+  total_timesteps:
+    value: 12000
+  eval_episodes:
+    value: 3
+  eval_freq:
+    value: 500
+  log_freq:
+    value: 100
+  revenue_weight:
+    value: 0.01
+  n_products:
+    value: 8
+  N:
+    value: 80
+  alpha:
+    value: 0.3
+  lambda_coi:
+    value: 0.2
+  robust_radius:
+    value: 0.0
+  robust_points:
+    value: 1
+  info_value:
+    value: 1.0
+  learning_rate:
+    value: 0.0003
+  gamma:
+    value: 0.99
+  buffer_size:
+    value: 20000
+  batch_size:
+    value: 128
+  tau:
+    value: 0.005
+  train_freq:
+    value: 1
+  learning_starts:
+    value: 500
+  n_steps:
+    value: 512
+  n_epochs:
+    value: 10
+  gae_lambda:
+    value: 0.95
+  clip_range:
+    value: 0.2
+  ent_coef:
+    value: 0.0
+  target_update_interval:
+    value: 500
+  exploration_fraction:
+    value: 0.2
+  exploration_final_eps:
+    value: 0.05
+  action_levels:
+    value: 7
+  action_scale_low:
+    value: 0.9
+  action_scale_high:
+    value: 1.1
+  q_lr:
+    value: 0.1
+  q_bins:
+    value: 6
+  eps_start:
+    value: 1.0
+  eps_end:
+    value: 0.05
+  eps_decay:
+    value: 0.9995

engine/sweeps/sac_tune.yaml

@@ -0,0 +1,54 @@
+method: bayes
+metric:
+  name: sweep/score
+  goal: maximize
+command:
+  - ${env}
+  - python
+  - -m
+  - engine.train
+parameters:
+  algo:
+    value: sac
+  total_timesteps:
+    values: [50000, 80000, 120000]
+  seed:
+    values: [13, 42, 77]
+  alpha:
+    distribution: uniform
+    min: 0.15
+    max: 0.55
+  n_products:
+    values: [8, 10, 12]
+  lambda_coi:
+    distribution: uniform
+    min: 0.05
+    max: 0.5
+  robust_radius:
+    distribution: uniform
+    min: 0.05
+    max: 0.3
+  robust_points:
+    values: [3, 5, 7]
+  info_value:
+    distribution: uniform
+    min: 0.5
+    max: 2.0
+  revenue_weight:
+    values: [0.005, 0.01, 0.02]
+  learning_rate:
+    distribution: log_uniform_values
+    min: 3.0e-5
+    max: 1.0e-3
+  gamma:
+    values: [0.98, 0.99, 0.995]
+  buffer_size:
+    values: [50000, 100000, 200000]
+  batch_size:
+    values: [128, 256, 512]
+  tau:
+    values: [0.002, 0.005, 0.01]
+  train_freq:
+    values: [1, 4, 8]
+  learning_starts:
+    values: [1000, 3000, 5000]

engine/sweeps/small_arch_compare.yaml

@@ -0,0 +1,86 @@
+method: random
+metric:
+  name: sweep/score
+  goal: maximize
+command:
+  - ${env}
+  - python
+  - -m
+  - engine.train
+parameters:
+  algo:
+    values: [ppo, a2c, dqn, qtable]
+  arch:
+    values: [tiny, small, medium]
+  activation:
+    values: [relu, tanh]
+  total_timesteps:
+    values: [8000, 12000, 20000]
+  seed:
+    values: [13, 42, 77]
+  n_products:
+    values: [6, 8, 10]
+  alpha:
+    distribution: uniform
+    min: 0.1
+    max: 0.5
+  lambda_coi:
+    distribution: uniform
+    min: 0.05
+    max: 0.4
+  robust_radius:
+    values: [0.0, 0.1, 0.2]
+  robust_points:
+    values: [3, 5]
+  info_value:
+    values: [0.75, 1.0, 1.5]
+  revenue_weight:
+    values: [0.005, 0.01, 0.02]
+  learning_rate:
+    distribution: log_uniform_values
+    min: 1.0e-5
+    max: 5.0e-4
+  gamma:
+    values: [0.98, 0.99]
+  buffer_size:
+    values: [10000, 30000, 50000]
+  batch_size:
+    values: [64, 128, 256]
+  tau:
+    values: [0.002, 0.005, 0.01]
+  train_freq:
+    values: [1, 4]
+  learning_starts:
+    values: [500, 1000, 2000]
+  n_steps:
+    values: [256, 512, 1024]
+  n_epochs:
+    values: [5, 10]
+  gae_lambda:
+    values: [0.9, 0.95]
+  clip_range:
+    values: [0.1, 0.2]
+  ent_coef:
+    values: [0.0, 0.005]
+  target_update_interval:
+    values: [500, 1000]
+  exploration_fraction:
+    values: [0.1, 0.2]
+  exploration_final_eps:
+    values: [0.02, 0.05]
+  action_levels:
+    values: [5, 7, 9]
+  action_scale_low:
+    values: [0.85, 0.9]
+  action_scale_high:
+    values: [1.1, 1.15]
+  q_lr:
+    values: [0.05, 0.1, 0.2]
+  q_bins:
+    values: [4, 6, 8]
+  eps_start:
+    value: 1.0
+  eps_end:
+    values: [0.02, 0.05]
+  eps_decay:
+    values: [0.999, 0.9995]