bacpipe.embedding_evaluation.visualization package

Submodules

bacpipe.embedding_evaluation.visualization.dashboard module

class bacpipe.embedding_evaluation.visualization.dashboard.DashBoard(model_names, audio_dir, main_results_dir, default_label_keys, evaluation_task, dim_reduction_model, dim_reduc_parent_dir, **kwargs)

Bases: DashBoardHelper

add_styling(*pages)

all_models_page(widget_idx)

apply_clfier_page(widget_idx)

build_layout()

Builds the layout for the dashboard with two models and a single model page. The layout consists of a single model page, a two-models comparison page, and a page showing all models. Each page contains sidebars with model-specific information and content areas for visualizations.

clustering_panel(widget_idx)

embedding_panel(widget_idx=0)

make_sidebar(widget_idx, model=True, classifier_page=False, all_models=False)

model_page(widget_idx, single_model=False)

probing_panel(widget_idx)

spectrogram_panel(widget_idx=0)

bacpipe.embedding_evaluation.visualization.dashboard.visualize_using_dashboard(models, dashboard_port=5006, dashboard_address='localhost', dashboard_websocket_origin=False, **kwargs)

Create and serve the dashboard for visualization. To colorcode embeddings by other labels than the default ones, create an annotations file with timestamps. An example file can be found in ‘bacpipe/tests/test_data/annotations.csv’. Multiple dashboards can be opened, the port will simply increment.

Parameters:

• models (list) – embedding models

• kwargs (dict) – Dictionary with parameters for dashboard creation

bacpipe.embedding_evaluation.visualization.dashboard_utils module

class bacpipe.embedding_evaluation.visualization.dashboard_utils.DashBoardHelper

Bases: object

add_save_button(plot_func, **kwargs)

Adds a save button to the plot panel

change_input_options(clfier_selection, widget_idx)

handle_click(event, widget_idx=0)

handle_selection(event, widget_idx=None)

Triggered when the user uses the Lasso or Box select tool.

init_interactive_embed_plot(widget_idx)

Initialize interactive embedding plot with dummy figure

init_plot(p_type, plot_func, widget_idx, **kwargs)

init_widget(idx, w_type, **kwargs)

plot_widget(plot_func, **kwargs)

save_embedding_plot(event, widget_idx, plot_func, **kwargs)

Save the current embedding plot

save_selected_points(event, dialogue_panel, widget_idx)

update_main_plot(p_type, plot_func, widget_idx, **kwargs)

Update existing plot by just updating the .object

widget(name, options, attr='Select', width=120, **kwargs)

bacpipe.embedding_evaluation.visualization.tooltips module

bacpipe.embedding_evaluation.visualization.visualize module

bacpipe.embedding_evaluation.visualization.visualize.clustering_overview(path_func, label_by, no_noise, model_list, label_column, **kwargs)

Create overview plots for clustering metrics.

Parameters:

• path_func (function) – function to return the paths when model name is given

• label_by (str) – key of default_labels dict

• no_noise (bool) – whether to plot the metrics with or without noise

• model_list (list) – list of models

• label_column (str) – label as defined in the annotations.csv file

• kwargs (dict) – additional arguments for plotting

Returns:

figure handle

Return type:

plt.plot object

bacpipe.embedding_evaluation.visualization.visualize.generate_bar_plot(metrics, fig, ax, x_label='Metric value', no_legend=False, **kwargs)

bacpipe.embedding_evaluation.visualization.visualize.iterate_through_subtasks(plot_func, plot_path, task_name, model_list, metrics)

For classification multiple subtasks exist (linear and knn). Iterate over each of the subtasks and call the plotting functions to create the visualizations.

Parameters:

• plot_func (function) – returns model specific paths when model name is passed

• plot_path (pathlib.Path object) – path to store overview plots

• task_name (str) – name of task

• model_list (list) – list of models

• metrics (dict) – performance dictionary

bacpipe.embedding_evaluation.visualization.visualize.plot_clusterings(path_func, model_name, label_by, no_noise, fig=None, ax=None, **kwargs)

Plot the clustering metrics for a given model and label type.

Parameters:

• path_func (function) – function to return the paths when model name is given

• model_name (str) – name of model

• label_by (str) – key of default_labels dict

• no_noise (bool) – whether to plot the metrics with or without noise

• fig (plt.plot object, optional) – figure handle, by default None

• ax (plt.plot object, optional) – axes handle, by default None

Returns:

figure handle

Return type:

plt.plot object

bacpipe.embedding_evaluation.visualization.visualize.plot_overview_metrics(plot_path, task_name, model_list, metrics, path_func=None, return_fig=False, sort_string='kmeans-audio_file_name')

Visualization of task performance by model accross all classes. Resulting plot is stored in the plot path.

Parameters:

• plot_path (pathlib.Path object) – path to store overview plots

• task_name (str) – name of task

• model_list (list) – list of models

• metrics (dict) – performance dictionary

• sort_string (str) – string to sort the metrics by, defaults to “kmeans-audio_file_name”

bacpipe.embedding_evaluation.visualization.visualize.visualise_results_across_models(plot_path, task_name, model_list)

Create visualizations to compare models by specified tasks.

Parameters:

• path_func (function) – return the paths when given a model name

• plot_path (pathlib.Path object) – path to overview plots

• task_name (str) – name of task

• model_list (list) – list of models

bacpipe.embedding_evaluation.visualization.visualize_embeddings module

class bacpipe.embedding_evaluation.visualization.visualize_embeddings.EmbedAndLabelLoader(dim_reduction_model, dashboard=False, **kwargs)

Bases: object

get_data(model_name, label_by, remove_noise=False, **kwargs)

remove_noise_indices(model_name)

bacpipe.embedding_evaluation.visualization.visualize_embeddings.collect_dim_reduced_embeds(model_name, dim_reduced_embed_path, dim_reduction_model, **kwargs)

Return the dimensionality reduced embeddings of a model.

Parameters:

• model_name (str) – name of model

• dim_reduced_embed_path (pathlib.Path object) – path to dim reduced embeddings

• dim_reduction_model (str) – name of feature extraction model

Returns:

dimensionality reduced embeddings

Return type:

dict

bacpipe.embedding_evaluation.visualization.visualize_embeddings.darken_hex_color_bitwise(hex_color)

Darkens a hex color using the bitwise operation: (color & 0xfefefe) >> 1.

Parameters:

hex_color (str): The hex color string (e.g., ‘#1f77b4’).

Returns:

str: The darkened hex color.

bacpipe.embedding_evaluation.visualization.visualize_embeddings.data_split_by_labels(embeds_dict, labels)

Split data by labels for scatterplots.

Parameters:

• embeds_dict (dict) – embeddings by model

• labels (list) – list of labels

Returns:

x and y data corresponding to labels

Return type:

dict

bacpipe.embedding_evaluation.visualization.visualize_embeddings.get_labels_for_plot(model_name=None, **kwargs)

bacpipe.embedding_evaluation.visualization.visualize_embeddings.init_embed_figure(fig, axes, bool_3d=False, **kwargs)

bacpipe.embedding_evaluation.visualization.visualize_embeddings.plot_comparison(plot_path, models, dim_reduction_model, bool_spherical=False, dashboard=False, loader=None, evaluation_task=[], **kwargs)

Create big overview visualization of all embeddings spaces. Labels are chosen from ground_truth and if that does not exist, default lables are used.

Parameters:

• plot_path (pathlib.Path object) – path to store overview plots

• models (list) – list of models

• dim_reduction_model (str) – name of dimensionality reduction model

• bool_spherical (bool, optional) – if True 3d embeddings will be plotted, by default False

• dashboard (bool, optional) – if dashboard called this function or not

• loader (EmbedAndLabelLoader object) – object containing embeds and labels by model for quicker loading

• evaluation_task (list, optional) – list of tasks to evaluate, by default []

Returns:

figure handle

Return type:

plt object

bacpipe.embedding_evaluation.visualization.visualize_embeddings.plot_embedding_points(axes, embeds, split_data, labels, c_label_dict, remove_noise=False, **kwargs)

Plot embeddings in scatter plot.

Parameters:

• axes (plt object) – axes handle

• embeds (dict) – embeddings

• split_data (dict) – data split by label

• labels (list) – labels of the data

• c_label_dict (dict) – linking labels to ints for coloring

• remove_noise (bool, optional) – remove noise or not, defaults to False

Returns:

axes points

Return type:

plt object

bacpipe.embedding_evaluation.visualization.visualize_embeddings.plot_embeddings(loader, model_name, label_by, paths=None, dim_reduction_model=None, axes=False, fig=False, dashboard=False, dashboard_idx=None, **kwargs)

Generate figures and axes to plot points corresponding to embeddings. This function can also be called and given figure and axes handeles. In that case the existing handles will be used to add the points and configure the axes and labels.

Parameters:

• loader (EmbedAndLabelLoader object) – contains the labels and embeddings by model, for quicker loading

• model_name (str) – name of model

• label_by (str, optional) – key of default_labels dict, by default “audio_file_name”

• paths (SimpleNamespace object, optional) – object with path attributes, defaults to None

• dim_reduction_model (str) – name of dim reduced model

• axes (plt object, optional) – axes handle, by default False

• fig (plt object, optional) – figure handle, by default False

• dashboard (bool, optional) – whether the calls comes from the dashboard, by deafult False

• dashboard_idx (int, optional) – index of dashboard plot, relevant for legend placement

Returns:

• plt object – axes handles is axes handles were given

• dict – color dictionary for legend

• list – plt point objects for legend of colorbar

bacpipe.embedding_evaluation.visualization.visualize_embeddings.plot_embeddings_px(embeds, labels, c_label_dict, label_by='label', **kwargs)

bacpipe.embedding_evaluation.visualization.visualize_embeddings.reorder_embeddings_by_clustering_performance(plot_path, axes, models, order_metric='ground_truth-kmeans')

Reorder the embedding overview plot by clustering performance.

Parameters:

• plot_path (pathlib.Path object) – path to store plots and results comparing all models

• axes (plt.axes object) – handle for figures axes

• models (list) – list of models

• order_metric (str) – key corresponding to a metric in the clustering_results.json file. Defaults to “ARI(kmeans)”

bacpipe.embedding_evaluation.visualization.visualize_embeddings.return_rows_cols(num)

bacpipe.embedding_evaluation.visualization.visualize_embeddings.set_colorbar_or_legend(fig, axes, points, c_label_dict, label_by, **kwargs)

bacpipe.embedding_evaluation.visualization.visualize_embeddings.set_figsize_for_comparison(rows, cols)

bacpipe.embedding_evaluation.visualization.visualize_embeddings.set_legend(handles, labels, fig, axes, bool_plot_centroids=False, dashboard=False, **kwargs)

Create the legend for embeddings visualization plots.

Parameters:

• handles (list) – list of legend handles

• labels (list) – list of labels for legend

• fig (plt.fig object) – figure handle

• axes (plt.axes object) – axes handle

• bool_plot_centroids (bool, optional) – if True centroids of each class will be plotted, by default True

• dashboard (bool) – if dashboard called this function or not

Returns:

• plt.fig object – figure handle

• plt.axes object – axes handle

bacpipe.embedding_evaluation.visualization.visualize_predictions module

class bacpipe.embedding_evaluation.visualization.visualize_predictions.PredictionsLoader(vis_loader, path_func, models, panel_selection, progress_bar, loading_pane, thresh=0.5)

Bases: object

accumulate_data(species, accumulate_by='day')

get_classes(path)

get_data(model, threshold, clfier_type=None, probe_path='', **kwargs)

get_timestamps_per_embedding(model)

load_classification(model, threshold)

static reorder_by_most_occurrance(probs, label2index)

static transform_presence_into_hour_heatmap(species_presence, hours, accumulator)

static verify_threshold(threshold)

bacpipe.embedding_evaluation.visualization.visualize_predictions.load_results(path_func, task, model_list)

Load the task results into a dict and return them. For classification multiple subtasks exist, so do them seperately.

Parameters:

• path_func (function) – returns model specific tasks when model is given

• task (str) – name of task

• model_list (list) – list of models

Returns:

performance for different tasks and models

Return type:

dict

bacpipe.embedding_evaluation.visualization.visualize_predictions.plot_classification_heatmap(event, predictions_loader, model, accumulate_by, threshold, species=None, **kwargs)

bacpipe.embedding_evaluation.visualization.visualize_predictions.plot_classification_results(task_name, paths=None, metrics=None, return_fig=False, path_func=None, model_name=None)

Save model specific classification results in the model specific plot path, displayed as horizontal bars.

Parameters:

• task_name (str) – name of task

• paths (SimpleNamespace object) – path to store plots

• metrics (dict) – classification performance

• return_fig (bool) – if True the figure will be returned, by default False

• path_func (function) – function to return the paths when model name is given

• model_name (str) – name of model, by default None

Returns:

figure handle

Return type:

plt object

bacpipe.embedding_evaluation.visualization.visualize_predictions.plot_per_class_metrics(plot_path, task_name, model_list, metrics)

Visualization of per class results. Resulting figure is stored in plot path. Models are sorted by the value of the first entry.

Parameters:

• plot_path (pathlib.Path object) – path to store plot in

• task_name (str) – name of task

• model_list (list) – list of models

• metrics (dict) – performance dictionary

bacpipe.embedding_evaluation.visualization.visualize_spectrograms module

class bacpipe.embedding_evaluation.visualization.visualize_spectrograms.SpectrogramPlot(audio_dir, loader, model_name, panel_static_text, **kwargs)

Bases: object

create_specs(audio)

static dummy_image(title)

load_audio(start, end, filename)

play_audio(event)

set_axis_lims_dep_sr(S_dB)

update_spectrogram(clickData=None, play_btn=None, autoplay_radio=None)

update_text(start_s, end_s, audiofilename, label)

Module contents