def __init__(
self,
window, n_multiv, n_kernels, w_kernel,
d_k, d_v, d_model, d_inner,
n_layers, n_head, drop_prob=0.1):
'''
Args:
window (int): the length of the input window size
n_multiv (int): num of univariate time series
n_kernels (int): the num of channels
w_kernel (int): the default is 1
d_k (int): d_model / n_head
d_v (int): d_model / n_head
d_model (int): outputs of dimension
d_inner (int): the inner-layer dimension of Position-wise Feed-Forward Networks
n_layers (int): num of layers in Encoder
n_head (int): num of Multi-head
drop_prob (float): the probability of dropout
'''
super(Single_Global_SelfAttn_Module, self).__init__()
self.window = window
self.w_kernel = w_kernel
self.n_multiv = n_multiv
self.d_model = d_model
self.drop_prob = drop_prob
self.conv2 = nn.Conv2d(1, n_kernels, (window, w_kernel))
self.in_linear = nn.Linear(n_kernels, d_model)
self.out_linear = nn.Linear(d_model, n_kernels)
self.layer_stack = nn.ModuleList([
EncoderLayer(d_model, d_inner, n_head, d_k, d_v, dropout=drop_prob)
for _ in range(n_layers)])
def __init__(self, hparams):
"""
Args:
window (int): the length of the input window size
n_multiv (int): num of univariate time series
n_kernels (int): the num of channels
w_kernel (int): the default is 1
d_k (int): d_model / n_head
d_v (int): d_model / n_head
d_model (int): outputs of dimension
d_inner (int): the inner-layer dimension of Position-wise Feed-Forward Networks
n_layers (int): num of layers in Encoder
n_head (int): num of Multi-head
drop_prob (float): the probability of dropout
"""
super(Single_Local_SelfAttn_Module, self).__init__()
self.hparams = hparams
self.window = hparams.window
self.w_kernel = hparams.w_kernel
self.n_multiv = hparams.n_multiv
self.d_model = hparams.d_model
self.drop_prob = hparams.drop_prob
self.conv1 = nn.Conv2d(1, hparams.n_kernels,
(hparams.local, hparams.w_kernel))
self.pooling1 = nn.AdaptiveMaxPool2d((1, hparams.n_multiv))
self.in_linear = nn.Linear(hparams.n_kernels, hparams.d_model)
self.out_linear = nn.Linear(hparams.d_model, hparams.n_kernels)
self.layer_stack = nn.ModuleList(
[EncoderLayer(hparams) for _ in range(hparams.n_layers)])