gents.evaluation.visualization package

Module contents

gents.evaluation.visualization.imputation_visual(real_data: Tensor, gen_data: Tensor, cond_data: Tensor, data_mask: BoolTensor, max_viz_n_channel=3, save_root=None)

Visualize time series imputation results, including 95% predict interval.

Parameters:

  • real_data (torch.Tensor) – Ground truth time series, in shape of [B, seq_len, C].

  • gen_data (torch.Tensor) – Predicted time series scenarios, in shape of [B, seq_len, C, N], N is the number of scenarios.

  • cond_data (torch.Tensor) – Observed time series, in shape of [B, seq_len, C], missing values should be set as NaN.

  • data_mask (torch.BoolTensor) – Ground truth time series data mask, in shape of [B, seq_len, C].

  • max_viz_n_channel (int, optional) – The maximum number of channels to be visualized. Defaults to 3.

  • save_root (str, optional) – Save root path. The post fix should be .png/.pdf/etc. If None, don’t save figure. Defaults to None.

gents.evaluation.visualization.predict_visual(real_data: Tensor, gen_data: Tensor, data_mask: BoolTensor, max_viz_n_channel: int = 3, save_root: str | None = None)

Visualize time series prediction results, including 95% predict interval.

Parameters:

  • real_data (torch.Tensor) – Ground truth time series, in shape of [B, obs_len + seq_len, C].

  • gen_data (torch.Tensor) – Predicted time series scenarios, in shape of [B, seq_len, C, N], N is the number of scenarios.

  • data_mask (torch.BoolTensor) – Ground truth time series data mask, in shape of [B, obs_len + seq_len, C].

  • max_viz_n_channel (int, optional) – The maximum number of channels to be visualized. Defaults to 3.

  • save_root (str, optional) – Save root path. The post fix should be .png/.pdf/etc. If None, don’t save figure. Defaults to None.

gents.evaluation.visualization.tsne_visual(real_data, generated_data, class_label_data=None, save_root=None, min_viz_samples=1000)

TSNE visualization of generated time series and real time series.

Parameters:

  • real_data (ArrayLike) – Real time series data, in shape of [B, T, C].

  • generated_data (ArrayLike) – Generated time series data, in shape of [B, T, C].

  • class_label_data (ArrayLike, optional) – Time series labels. If not None, in shape of [B, ] Defaults to None.

  • save_root (str, optional) – Save root path. The post fix should be .png/.pdf/etc. If None, don’t save figure. Defaults to None.

  • min_viz_samples (int, optional) – The number of data samples put into visualization. Defaults to 1000.