utils package

Submodules

utils.adapt_variables module

utils.adapt_variables.adapt_variables(config, x, masks, labels)

Adapts the variables to fit the dimension of the model selected

Parameters:

  • config (configuration dictionary.)

  • x (input data of dims)

  • masks (masks locating where we have values or not.)

  • labels (ground truth data.)

  • form (Inputs of the) –

    1D: (batch, variables, time)

    2D: (batch, variables, time, x)

    can be converted to: -> 1D: (batch * x, variables, time)

    3D: (batch, variables, time, lat, lon)

    can be converted to: -> 1D: (batch * (lat * lon), variables, time) -> 2D: (batch * time, variables, lat, lon)

Returns:

  • x (adapted input.)

  • masks (adapted masks.)

  • labels (adapted labels.)

utils.determine_threshold module

utils.determine_threshold.determine_threshold(y_pred, y_true, metric_name='f1_score', metric_params={}, num=100, lower_is_better=False)

Find optimal threshold based on defined metric.

Parameters:

  • y_pred (Array) – Prediction/output variables by your model.

  • y_true (Array) – Target/ground-truth labels.

  • metric_name (Str) – Torchmetrics for threshold optimization, defaults to “f1_score”.

  • metric_params (Dict) – Parameters for Torchmetrics, defaults to {}.

  • num (Int) – Number of candidates for threshold optimization, defaults to 100.

  • lower_is_better (Float) – Indicate whether or not lower metric value is better, defaults to False.

Returns:

Dictionary with metrics from torchmetrics to be evaluated.

Return type:

Dict

utils.loss module

class utils.loss.Loss(package, configuration)

Bases: Module

This class is used for wrapping for Torch loss functions.

forward(outputs, labels)

Forward function of the loss, playing the same role as __call__ does for a regular python class.

Parameters:

  • outputs (Torch.Tensor) – Prediction/output variables by your model.

  • labels (Torch.Tensor) – Target/ground-truth labels.

Returns:

The loss with respect to all Tensors.

Return type:

Torch.Tensor

utils.loss.set_loss(parameters)

Check if loss is user defined.

utils.metrics_pyod module

utils.metrics_pyod.evaluate_print(clf_name, y, scores, config, threshold=0.5)

Utility function for evaluating and printing the results obtained given a PyOD model. Default metrics include ROC, Accuracy, Precision, Recall and Average Precision.

Parameters:

  • clf_name (str) – Name for the experiment.

  • y (np.array) – Target/ground-truth labels.

  • scores (np.array) – Output scores provided by the model.

  • threshold (float, optional) – Threshold to compute binary scores for evaluation purposes, defaults to 0.5

utils.metrics_pytorch module

utils.metrics_pytorch.init_metrics(config)

Initialization of metric objects for evaluation purposes.

Parameters:

config (dict) – A dictionary containing all the configuration variables for the experiment.

Returns:

Dictionary with metrics from torchmetrics to be evaluated.

Return type:

dict

utils.misc module

utils.misc.plot_confusion_matrix(cm, outputs, labels, path)

Compute and plot confusion matrix for evaluation

Parameters:

  • cm (class: 'torchmetrics.ConfusionMatrix') – Confusion matrix object

  • outputs (tensor) – Categorized model output

  • labels (tensor) – Samples’ ground truth

  • path (str) – Path to save the generated figures

utils.misc.plot_spatial_aggregation(path, output, label, mask, coordinates, num_classes)

Plot extreme event detection saliency map for a given time step

Parameters:

  • path (str) – Path to save the generated figures

  • output – Categorized model output

  • label – Samples’ ground truth

  • mask – Quality mask

  • coordinates (tuple) – Bounding box of the region of interest (lon_min, lon_max, lat_min, lat_max)

  • num_classes (int) – Number of classess in the classification

utils.misc.plot_spatial_signal(path, output, label, mask, coordinates, contourf_params, title)

Plot extreme event detection categorical map for a given time step

Parameters:

  • path (str) – Path to save the generated figures

  • output – Decision scores at the output of the model

  • label – Samples’ ground truth

  • mask – Quality mask

  • coordinates (tuple) – Bounding box of the region of interest (lon_min, lon_max, lat_min, lat_max)

  • contourf_params (dict) – Dictionary containing customizable parameters for the contourf function

  • title (str) – Title of the generated visualization

utils.misc.plot_temporal_results(path, xticks_labels, labels, outputs_when_event, output_without_event=None)
Plot extreme event detection signals through time:

  • Ground truth extreme event signal

  • Extreme signal (Aggregated mean over region affected by the extreme event)

  • Non-extreme signal (Aggregated mean over region non-affected by the extreme event)

Parameters:

  • path (str) – Path to save the generated figures

  • xticks_labels (list) – X axis’ labels

  • labels (list) – Samples’s ground truth

  • outputs_when_event (list) – Aggregated mean over region affected by the extreme event

  • output_without_event (list, optional) – Aggregated mean over region non-affected by the extreme event, defaults to None

utils.setup_config module

utils.setup_config.setup(filename)

Opens the configuration file as a dictionary

Parameters:

filename (str) – path to .yaml configuration file

Returns:

configuration file

Return type:

dict

utils.setup_config.setup_config()

Gets the configuration for the current experiment and creates the folder structure. If provided, some hyperparameters can be added

Returns:

experiment configuration

Return type:

dict

utils.validate_config module

utils.validate_config.validate_config(config)

Checks for the consistency between the data dimension, the dimension of the model requested and the output dimension

Parameters:

config (dict) – configuration file

Module contents