= 16
bs vars = 3
= 12
seq_len = 6
c_out = torch.rand(bs, vars, seq_len)
xb vars,c_out)(xb).shape, [bs, c_out])
test_eq(XceptionTime(vars,c_out, bottleneck=False)(xb).shape, [bs, c_out])
test_eq(XceptionTime(vars,c_out, residual=False)(xb).shape, [bs, c_out])
test_eq(XceptionTime(3, 2)), 399540) test_eq(count_parameters(XceptionTime(
XceptionTime
This is an unofficial PyTorch implementation by Ignacio Oguiza - oguiza@timeseriesAI.co modified on:
Fawaz, H. I., Lucas, B., Forestier, G., Pelletier, C., Schmidt, D. F., Weber, J. & Petitjean, F. (2019). InceptionTime: Finding AlexNet for Time Series Classification. arXiv preprint arXiv:1909.04939.
Official InceptionTime tensorflow implementation: https://github.com/hfawaz/InceptionTime
XceptionTime
XceptionTime (c_in, c_out, nf=16, nb_filters=None, adaptive_size=50, residual=True)
Same as nn.Module
, but no need for subclasses to call super().__init__
XceptionBlock
XceptionBlock (ni, nf, residual=True, ks=40, bottleneck=True)
Same as nn.Module
, but no need for subclasses to call super().__init__
XceptionModule
XceptionModule (ni, nf, ks=40, bottleneck=True)
Same as nn.Module
, but no need for subclasses to call super().__init__
= XceptionTime(2,3)
m 0].sum(), 5) # 2 shortcut + 3 bn
test_eq(check_weight(m, is_bn)[len(check_bias(m, is_conv)[0]), 0)
test_eq(len(check_bias(m)[0]), 5) # 2 shortcut + 3 bn test_eq(
3, 2) XceptionTime(
XceptionTime(
(block): XceptionBlock(
(xception): ModuleList(
(0): XceptionModule(
(bottleneck): Conv1d(3, 16, kernel_size=(1,), stride=(1,), bias=False)
(convs): ModuleList(
(0): SeparableConv1d(
(depthwise_conv): Conv1d(16, 16, kernel_size=(39,), stride=(1,), padding=(19,), groups=16, bias=False)
(pointwise_conv): Conv1d(16, 16, kernel_size=(1,), stride=(1,), bias=False)
)
(1): SeparableConv1d(
(depthwise_conv): Conv1d(16, 16, kernel_size=(19,), stride=(1,), padding=(9,), groups=16, bias=False)
(pointwise_conv): Conv1d(16, 16, kernel_size=(1,), stride=(1,), bias=False)
)
(2): SeparableConv1d(
(depthwise_conv): Conv1d(16, 16, kernel_size=(9,), stride=(1,), padding=(4,), groups=16, bias=False)
(pointwise_conv): Conv1d(16, 16, kernel_size=(1,), stride=(1,), bias=False)
)
)
(maxconvpool): Sequential(
(0): MaxPool1d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv1d(3, 16, kernel_size=(1,), stride=(1,), bias=False)
)
(concat): Concat(dim=1)
)
(1): XceptionModule(
(bottleneck): Conv1d(64, 32, kernel_size=(1,), stride=(1,), bias=False)
(convs): ModuleList(
(0): SeparableConv1d(
(depthwise_conv): Conv1d(32, 32, kernel_size=(39,), stride=(1,), padding=(19,), groups=32, bias=False)
(pointwise_conv): Conv1d(32, 32, kernel_size=(1,), stride=(1,), bias=False)
)
(1): SeparableConv1d(
(depthwise_conv): Conv1d(32, 32, kernel_size=(19,), stride=(1,), padding=(9,), groups=32, bias=False)
(pointwise_conv): Conv1d(32, 32, kernel_size=(1,), stride=(1,), bias=False)
)
(2): SeparableConv1d(
(depthwise_conv): Conv1d(32, 32, kernel_size=(9,), stride=(1,), padding=(4,), groups=32, bias=False)
(pointwise_conv): Conv1d(32, 32, kernel_size=(1,), stride=(1,), bias=False)
)
)
(maxconvpool): Sequential(
(0): MaxPool1d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv1d(64, 32, kernel_size=(1,), stride=(1,), bias=False)
)
(concat): Concat(dim=1)
)
(2): XceptionModule(
(bottleneck): Conv1d(128, 64, kernel_size=(1,), stride=(1,), bias=False)
(convs): ModuleList(
(0): SeparableConv1d(
(depthwise_conv): Conv1d(64, 64, kernel_size=(39,), stride=(1,), padding=(19,), groups=64, bias=False)
(pointwise_conv): Conv1d(64, 64, kernel_size=(1,), stride=(1,), bias=False)
)
(1): SeparableConv1d(
(depthwise_conv): Conv1d(64, 64, kernel_size=(19,), stride=(1,), padding=(9,), groups=64, bias=False)
(pointwise_conv): Conv1d(64, 64, kernel_size=(1,), stride=(1,), bias=False)
)
(2): SeparableConv1d(
(depthwise_conv): Conv1d(64, 64, kernel_size=(9,), stride=(1,), padding=(4,), groups=64, bias=False)
(pointwise_conv): Conv1d(64, 64, kernel_size=(1,), stride=(1,), bias=False)
)
)
(maxconvpool): Sequential(
(0): MaxPool1d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv1d(128, 64, kernel_size=(1,), stride=(1,), bias=False)
)
(concat): Concat(dim=1)
)
(3): XceptionModule(
(bottleneck): Conv1d(256, 128, kernel_size=(1,), stride=(1,), bias=False)
(convs): ModuleList(
(0): SeparableConv1d(
(depthwise_conv): Conv1d(128, 128, kernel_size=(39,), stride=(1,), padding=(19,), groups=128, bias=False)
(pointwise_conv): Conv1d(128, 128, kernel_size=(1,), stride=(1,), bias=False)
)
(1): SeparableConv1d(
(depthwise_conv): Conv1d(128, 128, kernel_size=(19,), stride=(1,), padding=(9,), groups=128, bias=False)
(pointwise_conv): Conv1d(128, 128, kernel_size=(1,), stride=(1,), bias=False)
)
(2): SeparableConv1d(
(depthwise_conv): Conv1d(128, 128, kernel_size=(9,), stride=(1,), padding=(4,), groups=128, bias=False)
(pointwise_conv): Conv1d(128, 128, kernel_size=(1,), stride=(1,), bias=False)
)
)
(maxconvpool): Sequential(
(0): MaxPool1d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv1d(256, 128, kernel_size=(1,), stride=(1,), bias=False)
)
(concat): Concat(dim=1)
)
)
(shortcut): ModuleList(
(0): ConvBlock(
(0): Conv1d(3, 128, kernel_size=(1,), stride=(1,), bias=False)
(1): BatchNorm1d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): ConvBlock(
(0): Conv1d(128, 512, kernel_size=(1,), stride=(1,), bias=False)
(1): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(add): Add
(act): ReLU()
)
(head): Sequential(
(0): AdaptiveAvgPool1d(output_size=50)
(1): ConvBlock(
(0): Conv1d(512, 256, kernel_size=(1,), stride=(1,), bias=False)
(1): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(2): ConvBlock(
(0): Conv1d(256, 128, kernel_size=(1,), stride=(1,), bias=False)
(1): BatchNorm1d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(3): ConvBlock(
(0): Conv1d(128, 2, kernel_size=(1,), stride=(1,), bias=False)
(1): BatchNorm1d(2, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(4): GAP1d(
(gap): AdaptiveAvgPool1d(output_size=1)
(flatten): Flatten(full=False)
)
)
)