paddlets.models.representation.dl.cost
- class CoST(segment_size: int, sampling_stride: int = 1, optimizer_fn: ~typing.Callable[[...], ~paddle.optimizer.optimizer.Optimizer] = <class 'paddle.optimizer.momentum.Momentum'>, optimizer_params: ~typing.Dict[str, ~typing.Any] = {'learning_rate': 0.001}, callbacks: ~typing.List[~paddlets.models.common.callbacks.callbacks.Callback] = [], batch_size: int = 128, max_epochs: int = 10, verbose: int = 1, seed: ~typing.Union[None, int] = None, repr_dims: int = 320, hidden_dims: int = 64, num_layers: int = 10, queue_size: int = 256, temperature: float = 0.07, alpha: float = 0.0005)[source]
Bases:
ReprBaseModelCoST[1] is a time series representation model published in ICLR 2022, It is a new time series representation learning framework for long sequence time series forecasting, which applies the contrastive learning method to learn disentangled seasonal-trend representations. CoST comprises both time domain and frequency domain contrastive losses to learn discriminative trend and seasonal representations, respectively.
[1] Woo G, et al. “CoST: Contrastive Learning of Disentangled Seasonal-Trend Representations for Time Series Forecasting”, https://arxiv.org/pdf/2202.01575.pdf
- Parameters
segment_size (int) – The size of time series segment.
sampling_stride (int) – Sampling intervals between two adjacent samples.
optimizer_fn (Callable[..., Optimizer]) – Optimizer algorithm.
optimizer_params (Dict[str, Any]) – Optimizer parameters.
callbacks (List[Callback]) – Customized callback functions.
batch_size (int) – Number of samples per batch.
max_epochs (int) – Max epochs during training.
verbose (int) – Verbosity mode.
seed (int|None) – Global random seed.
repr_dims (int) – The dimension of representation.
hidden_dims (int) – The number of channels in the hidden layer.
num_layers (int) – The number of ConvLayer to be stacked.
queue_size (int) – The dynamic queue size for saving negative examples.
temperature (float) – The temperature coefficient.
alpha (float) – The weight of seasonal components in loss.
- _segment_size
The size of time series segment.
- Type
int
- _sampling_stride
Sampling intervals between two adjacent samples.
- Type
int
- _optimizer_fn
Optimizer algorithm.
- Type
Callable[…, Optimizer]
- _optimizer_params
Optimizer parameters.
- Type
Dict[str, Any]
- _batch_size
Number of samples per batch.
- Type
int
- _max_epochs
Max epochs during training.
- Type
int
- _verbose
Verbosity mode.
- Type
int
- _seed
Global random seed.
- Type
int|None
- _repr_dims
The dimension of representation.
- Type
int
The number of channels in the hidden layer.
- Type
int
- _num_layers
The number of ConvLayer to be stacked.
- Type
int
- _queue_size
The dynamic queue size for saving negative examples.
- Type
int
- _temperature
The temperature coefficient.
- Type
float
- _alpha
The parameter control the weightage of seasonal components in loss.
- Type
float