takuseno/minerva

export const Q_FUNC_TYPE_OPTIONS = {
  mean: 'MEAN',
  qr: 'QUANTILE REGRESSION',
  iqn: 'IMPLICIT QUANTILE NETWORK',
  fqf: 'FULLY PARAMETERIZED QUANTILE NETWORK'
}

export const SCALER_OPTIONS = {
  null: 'NONE',
  standard: 'STANDARDIZATION',
  min_max: 'MIN MAX',
  pixel: 'PIXEL'
}

export const COMMON_CONFIGS = {
  basic_config: {
    n_epochs: 100,
    n_steps_per_epoch: 1000,
    q_func_factory: 'mean',
    scaler: null,
    n_frames: 4,
    batch_size: 100,
    use_gpu: null
  },
  advanced_config: {
    n_steps: 1,
    gamma: 0.99,
    n_critics: 1
  }
}

export const CONTINUOUS_CONFIGS = {
  awac: {
    actor_learning_rate: 3e-4,
    critic_learning_rate: 3e-4,
    batch_size: 256,
    tau: 0.005,
    lam: 1.0,
    n_action_samples: 1,
    n_critics: 2
  },
  bcq: {
    actor_learning_rate: 1e-3,
    critic_learning_rate: 1e-3,
    imitator_learning_rate: 1e-3,
    batch_size: 100,
    tau: 0.005,
    update_actor_interval: 1,
    lam: 0.75,
    n_action_samples: 100,
    action_flexibility: 0.05,
    rl_start_step: 0,
    latent_size: 32,
    beta: 0.5,
    n_critics: 2
  },
  bear: {
    actor_learning_rate: 3e-4,
    critic_learning_rate: 3e-4,
    imitator_learning_rate: 1e-3,
    temp_learning_rate: 3e-4,
    alpha_learning_rate: 1e-3,
    batch_size: 100,
    tau: 0.005,
    update_actor_interval: 1,
    initial_temperature: 1.0,
    initial_alpha: 1.0,
    alpha_threshold: 0.05,
    lam: 0.75,
    n_action_samples: 100,
    n_target_samples: 10,
    n_mmd_action_samples: 4,
    mmd_sigma: 20.0,
    warmup_steps: 0,
    n_critics: 2
  },
  cql: {
    actor_learning_rate: 3e-5,
    critic_learning_rate: 3e-4,
    temp_learning_rate: 3e-5,
    alpha_learning_rate: 0.0,
    batch_size: 256,
    tau: 0.005,
    update_actor_interval: 1,
    initial_temperature: 1.0,
    initial_alpha: 1.0,
    conservative_weight: 5.0,
    alpha_threshold: 10.0,
    n_action_samples: 10,
    n_critics: 2,
    soft_q_backup: false
  },
  crr: {
    actor_learning_rate: 3e-4,
    critic_learning_rate: 3e-4,
    batch_size: 100,
    target_update_interval: 100,
    n_action_samples: 4
  },
  ddpg: {
    actor_learning_rate: 3e-4,
    critic_learning_rate: 3e-4,
    batch_size: 100,
    tau: 0.005
  },
  plas: {
    actor_learning_rate: 3e-4,
    critic_learning_rate: 3e-4,
    imitator_learning_rate: 3e-4,
    batch_size: 256,
    tau: 0.005,
    lam: 0.75,
    action_flexibility: 0.05,
    update_actor_interval: 1,
    warmup_steps: 10000,
    beta: 0.5,
    n_critics: 2
  },
  sac: {
    actor_learning_rate: 3e-4,
    critic_learning_rate: 3e-4,
    temp_learning_rate: 3e-4,
    batch_size: 256,
    tau: 0.005,
    n_critics: 2,
    update_actor_interval: 1,
    initial_temperature: 1.0
  },
  td3: {
    actor_learning_rate: 3e-4,
    critic_learning_rate: 3e-4,
    batch_size: 256,
    tau: 0.005,
    target_smoothing_sigma: 0.2,
    target_smoothing_clip: 0.5,
    update_actor_interval: 2,
    n_critics: 2
  }
}

export const DISCRETE_CONFIGS = {
  bcq: {
    learning_rate: 6.25e-5,
    batch_size: 32,
    action_flexibility: 0.3,
    beta: 0.5
  },
  cql: {
    learning_rate: 6.25e-5,
    batch_size: 32,
    target_update_interval: 8000
  },
  dqn: {
    learning_rate: 6.25e-5,
    batch_size: 32,
    target_update_interval: 8000
  },
  double_dqn: {
    learning_rate: 6.25e-5,
    batch_size: 32,
    target_update_interval: 8000
  },
  sac: {
    actor_learning_rate: 3e-4,
    critic_learning_rate: 3e-4,
    temp_learning_rate: 3e-4,
    batch_size: 64,
    initial_temperature: 1.0,
    target_update_interval: 8000,
    n_critics: 2
  }
}

// Scalar values
export const STATUS_API_CALL_INTERVAL = 5000
export const FETCH_EXPERIMENTS_INTERVAL = 5000
export const DISPLAY_DECIMAL_LENGTH = 2
export const GRAPH_DIMMED_OPACITY = 0.1