def __init__(self):
self.mid_dim = 14
self.num_class = 2
super().__init__()
self.fc0 = M.Linear(self.num_class, self.mid_dim, bias=True)
self.bn0 = M.BatchNorm1d(self.mid_dim)
self.fc1 = M.Linear(self.mid_dim, self.mid_dim, bias=True)
self.bn1 = M.BatchNorm1d(self.mid_dim)
self.fc2 = M.Linear(self.mid_dim, self.num_class, bias=True)
def __init__(self):
self.mid_dim = 14
self.num_class = 2
super().__init__()
self.fc0 = M.Linear(self.num_class, self.mid_dim, bias=True)
self.bn0 = M.BatchNorm1d(self.mid_dim)
self.fc1 = M.Linear(self.mid_dim, self.mid_dim, bias=True)
self.bn1 = M.BatchNorm1d(self.mid_dim)
self.fc2 = M.Linear(self.mid_dim, self.num_class, bias=True)
self.data = np.random.random((12, 2)).astype(np.float32)
def __init__(self, converter="normal"):
self.converter = converter
self.mid_dim = 14
self.num_class = 2
super().__init__()
self.fc0 = M.Linear(self.num_class, self.mid_dim, bias=True)
self.bn0 = M.BatchNorm1d(self.mid_dim)
self.fc1 = M.Linear(self.mid_dim, self.mid_dim, bias=True)
self.bn1 = M.BatchNorm1d(self.mid_dim)
self.fc2 = M.Linear(self.mid_dim, self.num_class, bias=True)
self.data = np.arange(24).reshape(12, 2).astype(np.float32)
def __init__(self, mode):
super().__init__()
self.mode = mode
self.data1 = np.random.random((1, 32, 32)).astype(np.float32)
self.data2 = np.random.random((20, 3, 24, 24)).astype(np.float32)
self.bn1d = M.BatchNorm1d(32)
self.bn2d = M.BatchNorm2d(3)
def __init__(self, transpose=False):
super().__init__()
self.transpose = transpose
self.data = np.random.random((10, 100)).astype(np.float32)
weight = np.random.random((200, 100) if transpose else (100, 200))
self.linear_weight = mge.Tensor(weight)
self.bn = M.BatchNorm1d(200)
def __init__(self, i, value_embedding, key_embedding):
self.key_embedding = key_embedding
super(TransformerBlock).__init__()
self.position_encoding = M.Linear(L, key_embedding)
self.init_map = M.Linear(i, key_embedding)
self.value_mapping = M.Linear(key_embedding, value_embedding)
self.key_mapping = M.Linear(key_embedding, key_embedding)
self.query_mapping = M.Linear(key_embedding, key_embedding)
self.norm = M.BatchNorm1d(key_embedding)
def __init__(self, gate_channel, reduction_ratio=16, num_layers=1):
super(ChannelGate, self).__init__()
gate_channels = [gate_channel]
gate_channels += [gate_channel // reduction_ratio] * num_layers
gate_channels += [gate_channel]
self.gate_c = M.Sequential(
Flatten(), M.Linear(gate_channels[0], gate_channels[1]),
M.BatchNorm1d(gate_channels[1]), M.ReLU(),
M.Linear(gate_channels[-2], gate_channels[-1]))
def __init__(self, input_size: int, output_size: int):
super(LinearBlock, self).__init__()
self.relu = M.ReLU()
self.normalize = M.BatchNorm1d(
output_size,
affine=True,
momentum=0.999,
eps=1e-3,
track_running_stats=False,
)
self.linear = M.Linear(input_size, output_size)
fc_init_(self.linear)
def __init__(self, feature_dim, channel, size=7):
"""initialzation
Args:
feature_dim (int): dimension number of output embedding
channel (int): channel number of input feature map
size (int, optional): size of input feature map. defaults to 7
"""
super().__init__()
self.size = size
self.bn1 = M.BatchNorm2d(channel)
self.dropout = M.Dropout(drop_prob=0.1)
self.fc = M.Linear(channel, feature_dim)
self.bn2 = M.BatchNorm1d(feature_dim, affine=False)
def __init__(self):
super().__init__()
self.data = np.random.random((10, 100)).astype(np.float32)
self.linear = M.Linear(100, 200, bias=False)
self.bn = M.BatchNorm1d(200)