c.LR: 1e-3,
},
},
c.ALPHA: {
c.OPTIMIZER: torch.optim.Adam,
c.KWARGS: {
c.LR: 1e-3,
},
},
},
# SAC
c.ACCUM_NUM_GRAD: 1,
c.BATCH_SIZE: 256,
c.BUFFER_WARMUP: 1000,
c.EVALUATION_PREPROCESSING: gt.Identity(),
c.GAMMA: 0.99,
c.LEARN_ALPHA: True,
c.MAX_GRAD_NORM: 10,
c.NUM_GRADIENT_UPDATES: 1,
c.NUM_PREFETCH: 1,
c.REWARD_SCALING: 1.,
c.STEPS_BETWEEN_UPDATE: 1,
c.TARGET_ENTROPY: -3.,
c.TARGET_UPDATE_INTERVAL: 1,
c.TAU: 0.005,
c.TRAIN_PREPROCESSING: gt.Identity(),
c.UPDATE_NUM: 0,
# Progress Tracking
c.CUM_EPISODE_LENGTHS: [0],
c.MODEL_ARCHITECTURE: FixedScheduler,
c.KWARGS: {
c.INTENTION_I: 0,
c.NUM_TASKS: num_tasks,
},
c.SCHEDULER_PERIOD: c.MAX_INT,
},
},
# DrQ
c.K:
2,
c.M:
2,
c.EVALUATION_PREPROCESSING:
gt.Identity(),
c.TRAIN_PREPROCESSING:
gt.Identity(),
# SAC
c.ACCUM_NUM_GRAD:
1,
c.BATCH_SIZE:
256,
c.BUFFER_WARMUP:
1000,
c.GAMMA:
0.99,
c.INITIAL_ALPHA:
1.,
c.LEARN_ALPHA:
