timeseries dataset with partial crops.

ConcatDataset that handles id mapping correctly (to allow to aggregate predictions)

Create a sequence of convolutional (`ni` to `nf`), ReLU (if `use_activ`) and `norm_type` layers.

Create a sequence of convolutional (`ni` to `nf`), ReLU (if `use_activ`) and `norm_type` layers.

basic conv1d

basic conv1d

extend nn.Conv1d to support SAME padding input: (n_sample, in_channels, n_length) output: (n_sample, out_channels, (n_length+stride-1)//

Resample on the fly

Convert ndarrays in sample to Tensors.

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

Flatten `x` to a single dimension, often used at the end of a model. `full` for rank-1 tensor

Normalize using given stats.

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

Flatten `x` to a single dimension, often used at the end of a model. `full` for rank-1 tensor

Input: X: (n_samples, n_channel, n_length) Y: (n_samples) Output: out: (n_samples) params:

Transforms sequence-level to sample-level labels majority vote: pick the most frequent label as segment label (i.e. suitable for single-label clas

Transforms data using a given function i.e. data_new = func(data) for input is True else label_new = func(label)

Layer that concats `AdaptiveAvgPool1d` and `AdaptiveMaxPool1d`.

Layer that concats `AdaptiveAvgPool1d` and `AdaptiveMaxPool1d`.

Layer that concats `AdaptiveAvgPool1d` and `AdaptiveMaxPool1d`.

Basic Block: conv1 -> convk -> conv1 params: in_channels: number of input channels out_channels: number of output c

Basic Stage: block_1 -> block_2 -> ... -> block_M

Select certain channels. axis: axis index of the channel axis

composes several transformations into a single one (as provided by torchvision.transforms.Compose)

Paper: https://arxiv.org/abs/2410.04133 Code: https://github.com/PKUDigitalHealth/ECGFounder Checkpoints: https://huggingface

Paper: https://arxiv.org/abs/2410.08559 Code: https://github.com/sehunfromdaegu/ECG_JEPA Checkpoints: https://drive.google.co

Paper: https://doi.org/10.1016/j.xcrm.2024.101875 Code: https://github.com/control-spiderman/ECGFM-KED Checkpoints: https://z

MLP Module with GELU activation fn + dropout.

MLP Module with GELU activation fn + dropout.

Create a forward hook on module `m`

Wrapper around SSKernel that generates A, B, C, dt according to HiPPO arguments. The SSKernel is expected to support the interface forward()

Wrapper around SSKernel that generates A, B, C, dt according to HiPPO arguments. The SSKernel is expected to support the interface forward()

inception time architecture

V-JEPA (https://openreview.net/forum?id=WFYbBOEOtv) Learnable Query Attention Pooling

Paper: https://arxiv.org/abs/2403.06659 Code: https://github.com/cheliu-computation/MERL-ICML2024 Checkpoints: https://drive.

extend nn.MaxPool1d to support SAME padding params: kernel_size: kernel size stride: the stride of the window. Default value

Robust MLFLowLogger that does not crash if connection is down

Stores a representation of and computes the SSKernel function K_L(A^dt, B^dt, C) corresponding to a discretized state space, where A is Normal + Low R

Stores a representation of and computes the SSKernel function K_L(A^dt, B^dt, C) corresponding to a discretized state space, where A is Normal + Low R

A sequence of TimeSeriesEncoders (somewhat similar to nn.Sequential) Note: this is the only module that gets instantiated directly in template_mod

squeeze excite block as used for example in LSTM FCN

squeeze excite block as used for example in LSTM FCN

Paper: https://arxiv.org/abs/2402.09450 Code: https://github.com/bakqui/ST-MEM Checkpoints: https://drive.google.com/file/d/1

helper module: transpose operation that can be used like a nn.module