Python get_base_ppo_conf 예제들

예제 #1

def get_conf():
    lr = 3e-5

    return {
        **get_base_ppo_conf(num_workers=10),
        'lr': lr,
        'env_config': ENV_CONF_COMMUNISM,
    }

예제 #2

def get_conf():
    lr = 1e-4

    return {
        **get_base_ppo_conf(num_workers=10),
        "lr_schedule": [
            [0, lr],
            [10_000_000, lr],
            [15_000_000, 0],
        ],
    }

예제 #3

def get_conf():
    lr = 1e-4

    return {
        **get_base_ppo_conf(num_workers=10), 'lr_schedule': [
            [0, lr],
            [10_000_000, lr],
            [15_000_000, 0],
        ],
        'framework':
        'torch'
    }

예제 #4

파일: train_lstm.py 프로젝트: akirasosa/aie-train

def get_conf():
    return {
        **get_base_ppo_conf(num_workers=10),
        "sgd_minibatch_size": 3000,  # 60 * 200 * 4 / 3000 = 16 steps of (B=60, L=50, dim)

        "lr": 3e-4,

        "multiagent": {
            "policies_to_train": ["learned"],
            "policies": {
                "learned": (None, OBS_SPACE_AGENT, ACT_SPACE_AGENT, {
                    "model": {
                        "custom_model": "my_model",
                        'max_seq_len': 50,
                    },
                }),
            },
            "policy_mapping_fn": lambda x: 'learned',
        },
    }

예제 #5

def get_conf():
    return {
        **get_base_ppo_conf(num_workers=10),
        'vf_loss_coeff': 0.1,
        'env_config': ENV_CONF_DYSTOPIA,
    }