Skip to content

Visualisations

DatasetMediaWriter

Write static and video media for the variables in downloaded datasets.

base_path = base_path instance-attribute

static(*, dataset, timestep)

Save static plots for one timestep of a downloaded dataset.

video(*, dataset, n_steps, timestep)

Save one animation per variable for a run of consecutive timesteps.

MediaPublisher

Publish static and video plots for a dataset/plot_spec/land_mask context.

idx_date = plot_spec.selected_timestep instance-attribute

panel_renderer = PanelRenderer(land_mask, replace(training_metadata, model=model_name, trained_epochs=trained_epochs), plot_spec) instance-attribute

log_static_inputs(inputs, dates, image_loggers, *, prefix=None)

Extract and log static raw input plots.

Parameters:

Name Type Description Default
inputs list[SingleDataset]

List of SingleDataset instances containing the input data.

 required
dates list[datetime]

List of datetime objects corresponding to the timesteps in the datasets.

 required
image_loggers list[SupportsImageLogging]

List of image loggers to send the rendered images to.

 required
prefix str | None

Optional prefix for the log path to namespace the logged images.

 None

log_static_outputs(outputs, image_loggers, *, channel_names=None, climatology=None, compare_truth_climatology=False, forecast_dates, history_dates, prefix=None, uncertainties=None)

Create and log static output plots, including climatology when available.

When a matching entry is present, also logs a standardised uncertainty (z-score) plot and a calendar-day-mean (climatology) map for the plotted date and channel.

Parameters:

Name Type Description Default
outputs ModelStepOutput

ModelStepOutput containing the ground truth and prediction arrays.

 required
image_loggers list[SupportsImageLogging]

List of image loggers to send the rendered images to.

 required
channel_names list[str] | None

Optional list of channel names corresponding to the output channels.

 None
climatology ArrayTCHW | None

Optional climatology array for the output channels.

 None
compare_truth_climatology bool

Whether to also log a ground-truth/climatology/ difference triplet, with a plain "on " title and no footer since it doesn't depend on the model. Callers should only set this once per date rather than every epoch.

 False
forecast_dates list[datetime]

List of forecast dates corresponding to the output timesteps.

 required
history_dates list[datetime]

List of history dates corresponding to the output timesteps.

 required
prefix str | None

Optional prefix for the log path to namespace the logged images.

 None
uncertainties dict[int, ArrayTHW] | None

Optional dictionary mapping channel indices to uncertainty arrays for the output channels.

 None

log_video_inputs(inputs, dates, video_loggers, *, prefix=None)

Extract and log raw input videos.

Parameters:

Name Type Description Default
inputs list[SingleDataset]

List of SingleDataset instances containing the input data.

 required
dates list[datetime]

List of datetime objects corresponding to the timesteps in the datasets.

 required
video_loggers list[SupportsVideoLogging]

List of video loggers to send the rendered videos to.

 required
prefix str | None

Optional prefix for the log path to namespace the logged videos.

 None

log_video_outputs(outputs, video_loggers, *, channel_names=None, climatology=None, forecast_dates, history_dates, compare_truth_climatology=False, prefix=None, uncertainties=None)

Create and log output videos.

Parameters:

Name Type Description Default
outputs ModelStepOutput

ModelStepOutput containing the ground truth and prediction arrays.

 required
video_loggers list[SupportsVideoLogging]

List of video loggers to send the rendered videos to.

 required
channel_names list[str] | None

Optional list of channel names corresponding to the output channels.

 None
climatology ArrayTCHW | None

Optional climatology array for the output channels.

 None
forecast_dates list[datetime]

List of forecast dates corresponding to the output timesteps.

 required
history_dates list[datetime]

List of history dates corresponding to the output timesteps.

 required
compare_truth_climatology bool

Whether to also log a ground-truth/climatology/ difference triplet, with a plain "on " title and no footer since it doesn't depend on the model. Callers should only set this once per date rather than every epoch.

 False
prefix str | None

Optional prefix for the log path to namespace the logged videos.

 None
uncertainties dict[int, ArrayTHW] | None

Optional dictionary mapping channel indices to uncertainty arrays for the output channels.

 None

register_animation_backends()

Register the ImageIO FFMPEG animation backend.