Implementation of TSEM: Temporally-Weighted Spatiotemporal Explainable Neural Network for Multivariate Time Series, in Pytorch
$ pip install tsemimport torch
from tsem.model import TSEM
series = torch.randn(1, 1, 256, 3)
window_size = series.shape[-2]
if window_size % 2 == 0:
window_size -= 1
tsem = TSEM(
window_size = window_size,
time_length = series.shape[-2],
feature_length = series.shape[-1],
n_classes = 1000
)
preds = tsem(series) # (1, 1000)window_size: int.
Window size along the time dimension. Must be smaller than time_length and odd number.time_length: int.
Length of the series.feature_length: int.
Number of series in the multivariate timeseries.num_classes: int.
Number of classes to classify.
- Training
criterion = nn.CrossEntropyLoss()
optimizer_ft = torch.optim.Adam(model.parameters(), lr=1.5e-4)
exp_lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer_ft, step_size=10, gamma=0.1)
model = train_model(model, criterion, optimizer_ft, exp_lr_scheduler, dataloaders, datasets_size, 50)- Explaining
from tsem.feature_extractors.cams import XGradCAM
from tsem.utils.visualization import CAMFeatureMaps
extracting_model = model
extracting_module = model.cnn_layers1_b1
targeting_layer = "relu_12"
feature_maps = CAMFeatureMaps(XGradCAM)
feature_maps.load(extracting_model,extracting_module,targeting_layer,use_cuda=True)
mask1 = feature_maps.show(X_test[0], True, None, plot=True)
feature_maps.map_activation_to_input(mask1[0])- Further details, please have a look in examples/
@inproceedings{pham2023tsem,
title={TSEM: Temporally-Weighted Spatiotemporal Explainable Neural Network for Multivariate Time Series},
author={Pham, Anh-Duy and Kuestenmacher, Anastassia and Ploeger, Paul G},
booktitle={Advances in Information and Communication: Proceedings of the 2023 Future of Information and Communication Conference (FICC), Volume 2},
pages={183--204},
year={2023},
organization={Springer}
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent