How to use the parl.layers.fc function in parl

def __init__(self, act_dim):
        self._act_dim = act_dim
        self._fc_1 = layers.fc(size=512, act='relu')
        self._fc_2 = layers.fc(size=256, act='relu')
        self._fc_3 = layers.fc(size=128, act='tanh')
        self._output = layers.fc(size=act_dim)

def __init__(self, act_dim):
        self.act_dim = act_dim
        #网络的层数、每层宽度均可微调
        self.fc0=layers.fc(size=20,act='tanh') 
        self.fc1=layers.fc(size=20,act='relu')
        self.fc = layers.fc(size=act_dim)

def __init__(self, act_dim):
        self._act_dim = act_dim
        self._fc_1 = layers.fc(size=512, act='relu')
        self._fc_2 = layers.fc(size=256, act='relu')
        self._fc_3 = layers.fc(size=128, act='tanh')
        self._output = layers.fc(size=act_dim)

self.vel_obs_dim = vel_obs_dim

        # buttom layers
        if shared:
            scope_name = 'policy_shared'
        else:
            scope_name = 'policy_identity_{}'.format(model_id)
        if stage_name is not None:
            scope_name = '{}_{}'.format(stage_name, scope_name)

        self.fc0 = layers.fc(
            size=hid0_size,
            act='tanh',
            param_attr=ParamAttr(name='{}/h0/W'.format(scope_name)),
            bias_attr=ParamAttr(name='{}/h0/b'.format(scope_name)))
        self.fc1 = layers.fc(
            size=hid1_size,
            act='tanh',
            param_attr=ParamAttr(name='{}/h1/W'.format(scope_name)),
            bias_attr=ParamAttr(name='{}/h1/b'.format(scope_name)))
        self.vel_fc0 = layers.fc(
            size=vel_hid0_size,
            act='tanh',
            param_attr=ParamAttr(name='{}/vel_h0/W'.format(scope_name)),
            bias_attr=ParamAttr(name='{}/vel_h0/b'.format(scope_name)))
        self.vel_fc1 = layers.fc(
            size=vel_hid1_size,
            act='tanh',
            param_attr=ParamAttr(name='{}/vel_h1/W'.format(scope_name)),
            bias_attr=ParamAttr(name='{}/vel_h1/b'.format(scope_name)))

        # top layers

def __init__(self, act_dim, algo='DQN'):
        self.act_dim = act_dim

        self.conv1 = layers.conv2d(
            num_filters=32, filter_size=5, stride=1, padding=2, act='relu')
        self.conv2 = layers.conv2d(
            num_filters=32, filter_size=5, stride=1, padding=2, act='relu')
        self.conv3 = layers.conv2d(
            num_filters=64, filter_size=4, stride=1, padding=1, act='relu')
        self.conv4 = layers.conv2d(
            num_filters=64, filter_size=3, stride=1, padding=1, act='relu')

        self.algo = algo
        if algo == 'Dueling':
            self.fc1_adv = layers.fc(size=512, act='relu')
            self.fc2_adv = layers.fc(size=act_dim)
            self.fc1_val = layers.fc(size=512, act='relu')
            self.fc2_val = layers.fc(size=1)
        else:
            self.fc1 = layers.fc(size=act_dim)

def __init__(self, act_dim):
        self._act_dim = act_dim
        self._fc_1 = layers.fc(size=512, act='relu')
        self._fc_2 = layers.fc(size=256, act='relu')
        self._fc_3 = layers.fc(size=128, act='tanh')
        self._output = layers.fc(size=act_dim)

def __init__(self, act_dim):
        hid1_size = 256
        hid2_size = 256

        self.fc1 = layers.fc(size=hid1_size, act='tanh')
        self.fc2 = layers.fc(size=hid2_size, act='tanh')
        self.fc3 = layers.fc(size=act_dim)

if shared:
            scope_name = 'critic_shared'
        else:
            scope_name = 'critic_identity_{}'.format(model_id)

        self.fc0 = layers.fc(
            size=hid0_size,
            act='selu',
            param_attr=ParamAttr(name='{}/w1/W'.format(scope_name)),
            bias_attr=ParamAttr(name='{}/w1/b'.format(scope_name)))
        self.fc1 = layers.fc(
            size=hid1_size,
            act='selu',
            param_attr=ParamAttr(name='{}/h1/W'.format(scope_name)),
            bias_attr=ParamAttr(name='{}/h1/b'.format(scope_name)))
        self.vel_fc0 = layers.fc(
            size=vel_hid0_size,
            act='selu',
            param_attr=ParamAttr(name='{}/vel_h0/W'.format(scope_name)),
            bias_attr=ParamAttr(name='{}/vel_h0/b'.format(scope_name)))
        self.vel_fc1 = layers.fc(
            size=vel_hid1_size,
            act='selu',
            param_attr=ParamAttr(name='{}/vel_h1/W'.format(scope_name)),
            bias_attr=ParamAttr(name='{}/vel_h1/b'.format(scope_name)))
        self.act_fc0 = layers.fc(
            size=hid1_size,
            act='selu',
            param_attr=ParamAttr(name='{}/a1/W'.format(scope_name)),
            bias_attr=ParamAttr(name='{}/a1/b'.format(scope_name)))

        # top layers

def __init__(self, act_dim):

        self.conv1 = layers.conv2d(
            num_filters=32, filter_size=8, stride=4, padding=1, act='relu')
        self.conv2 = layers.conv2d(
            num_filters=64, filter_size=4, stride=2, padding=2, act='relu')
        self.conv3 = layers.conv2d(
            num_filters=64, filter_size=3, stride=1, padding=0, act='relu')

        self.fc = layers.fc(size=512, act='relu')

        self.policy_fc = layers.fc(size=act_dim)
        self.value_fc = layers.fc(size=1)

def __init__(self, obs_dim, act_dim):
        super(ValueModel, self).__init__()
        hid1_size = obs_dim * 10
        hid3_size = 5
        hid2_size = int(np.sqrt(hid1_size * hid3_size))

        self.lr = 1e-2 / np.sqrt(hid2_size)

        self.fc1 = layers.fc(size=hid1_size, act='tanh')
        self.fc2 = layers.fc(size=hid2_size, act='tanh')
        self.fc3 = layers.fc(size=hid3_size, act='tanh')
        self.fc4 = layers.fc(size=1)