detoxai.visualization

Submodules

detoxai.visualization.DataVisualizer submodule

class detoxai.visualization.DataVisualizer.DataVisualizer(data_loader: DetoxaiDataLoader, plot_config: dict = {}, draw_rectangles: bool = False, rectangle_config: dict = {})

Bases: ImageVisualizer

visualize_batch(batch_num: int, max_images: int | None = 36, batch_preds: Tensor | None = None, show_labels: bool = True) → None

Parameters:

• batch_num – int:

• max_images – int | None: (Default value = 36)

• batch_preds – torch.Tensor | None: (Default value = None)

• show_labels – bool: (Default value = True)

Returns:

visualize_agg(batch_num: int) → None

Parameters:

batch_num – int:

Returns:

detoxai.visualization.HeatmapVisualizer submodule

class detoxai.visualization.HeatmapVisualizer.HeatmapVisualizer(data_loader: DetoxaiDataLoader | DataLoader, model: Module, lrp_object: LRPHandler = None, plot_config: dict = {}, draw_rectangles: bool = False, rectangle_config: dict = {})

Bases: ImageVisualizer

visualize_batch(batch_num: int, condition_on: ConditionOn = ConditionOn.PROPER_LABEL, show_cbar: bool = True, max_images: int | None = 36, return_fig: bool = False) → None

Parameters:

• batch_num – int:

• condition_on – ConditionOn: (Default value = ConditionOn.PROPER_LABEL)

• show_cbar – bool: (Default value = True)

• max_images – int | None: (Default value = 36)

• return_fig – bool: (Default value = False)

Returns:

visualize_agg(batch_num: int, condition_on: ConditionOn = ConditionOn.PROPER_LABEL) → None

Parameters:

• batch_num – int:

• condition_on – ConditionOn: (Default value = ConditionOn.PROPER_LABEL)

Returns:

detoxai.visualization.ImageVisualizer submodule

class detoxai.visualization.ImageVisualizer.ImageVisualizer

Bases: Visualizer, ABC

abstractmethod visualize_batch(batch_num: int) → None

Parameters:

batch_num – int:

Returns:

abstractmethod visualize_agg(batch_num: int) → None

Parameters:

batch_num – int:

Returns:

init_rectangle_painter(draw_rectangles: bool, rectangle_config: dict)

Initialize the rectangle painter.

Parameters:

• draw_rectangles (bool) – Whether to draw rectangles on the image.

• rectangle_config (dict) – The configuration for the rectangles.

Should contain the following keys: rect (tuple): The rectangle coordinates in the form (x, y, width, height). Mandatory. color (str): The color of the rectangle. Default is “black”. linewidth (int): The width of the rectangle’s border. Default is 2.

draw_rectangles: bool: rectangle_config: dict:

Returns:

maybe_paint_rectangle(ax) → None

Maybe paint a rectangle on the image. The rectangle is painted only if the draw_rectangles flag is set to True. Before calling this method, the init_rectangle_painter method should be called.

Parameters:

ax

Returns:

detoxai.visualization.LRPHandler submodule

class detoxai.visualization.LRPHandler.LRPHandler(attributor_name: str = 'Gradient', composite_name: str = 'EpsilonPlus', canonizers: list[str] = [], n_classes: int | None = None, **kwargs)

Bases: object

LRPHandler is a class that handles the calculation of input image attributions for a given model and dataset.

calculate(model: Module, data_loader: object, batch_num: int | None, max_images: int | None = None) → Tensor

Calculate LRP attribution

Parameters:

• model – nn

• data_loader – WrappedDataLoader | torch Dataloader

• batch_num – int

• attribution – is calculated for the entire datset

• model – nn.Module:

• data_loader – object:

• batch_num – int | None:

• max_images – int | None: (Default value = None)

Returns:

• torch.Tensor of size (L, N, IMG_w, IMG_h), where N is the number of samples, L is the class on which

LRP was conditioned, IMG_w and IMG_h are the width and height of the image.

detoxai.visualization.MetricsVisualizer submodule

class detoxai.visualization.MetricsVisualizer.MetricsVisualizer

Bases: Visualizer, ABC

metrics_config: dict | None = None

classmethod canonize_results(results: dict, metrics_config: dict) → DataFrame

Parameters:

• results – dict:

• metrics_config – dict:

Returns:

detoxai.visualization.MetricsVisualizer.results_to_tidy_df(results, metrics_config)

Parameters:

• results

• metrics_config

Returns:

detoxai.visualization.SSVisualizer submodule

class detoxai.visualization.SSVisualizer.SSVisualizer(data_loader: DetoxaiDataLoader | DataLoader, model: Module, lrp_object: LRPHandler = None, plot_config: dict = {}, draw_rectangles: bool = False, rectangle_config: dict = {})

Bases: ImageVisualizer

SS - Side by Side visualizer for Heatmaps and Data

visualize_batch(batch_num: int, lrp_condition_on: ConditionOn = ConditionOn.PROPER_LABEL, lrp_show_cbar: bool = True, max_images: int | None = 36, show_labels: bool = True) → None

Parameters:

• batch_num – int:

• lrp_condition_on – ConditionOn: (Default value = ConditionOn.PROPER_LABEL)

• lrp_show_cbar – bool: (Default value = True)

• max_images – int | None: (Default value = 36)

• show_labels – bool: (Default value = True)

Returns:

visualize_agg(batch_num: int) → None

Parameters:

batch_num – int:

Returns:

detoxai.visualization.ScatterVisualizer submodule

class detoxai.visualization.ScatterVisualizer.ScatterVisualizer(plots_config: dict = {})

Bases: Visualizer

create_plot(results: DataFrame | dict[str, CorrectionResult], rows: list[str] = ['Accuracy', 'GMean', 'F1'], cols: list[str] = ['Equalized_odds', 'Demographic_parity', 'Equal_opportunity', 'Accuracy_parity'], ticks_x: int = 4, ticks_y: int = 4, round_to: int = 3)

Parameters:

• results – pd.DataFrame | dict[str, CorrectionResult]

• rows – list[str]: (Default value = [“Accuracy”)

• "GMean"

• "F1"]

• cols – list[str]: (Default value = [“Equalized_odds”)

• "Demographic_parity"

• "Equal_opportunity"

• "Accuracy_parity"

• ]

• ticks_x – int: (Default value = 4)

• ticks_y – int: (Default value = 4)

• round_to – int: (Default value = 3)

Returns:

detoxai.visualization.TableVisualizer submodule

class detoxai.visualization.TableVisualizer.TableVisualizer(results: dict, metrics_config: dict)

Bases: MetricsVisualizer

create_table(*args, **kwargs)

get_table() → DataFrame

Get formated table aggregated for all methods

get_latex_table() → str

Get formatted table in LaTeX format with enhanced styling

get_raw_results() → DataFrame

Get raw results, i.e., the results as a melted DataFrame

save_to_csv(path: str) → None

Save the table to a CSV file

Parameters:

path – str:

Returns:

detoxai.visualization.Visualizer submodule

class detoxai.visualization.Visualizer.Visualizer

Bases: ABC

Base class for visualizers with plotting and logging capabilities.

This class provides basic functionality for creating, displaying, saving, and logging plots using matplotlib. It can be used with an ExperimentLogger for experiment tracking.

set_up_plots_configuration(config: dict) → None

Configure plotting parameters from a configuration dictionary.

Parameters:

config – Dictionary containing plot configuration parameters: - plt_save_kwargs: Dict of kwargs for plt.savefig (default: {}) - plot_style: Matplotlib style to use (default: ‘ggplot’) - shape_multiplier: Multiplier for plot dimensions (default: 1) - fontsize: Base font size for plots (default: 12)

attach_logger(logger: ExperimentLogger) → None

Attach a logger to the visualizer, enabling logging of plots

Parameters:

logger – ExperimentLogger:

show_plot() → None

Display the current figure using matplotlib’s show method.

This method displays the figure stored in self.figure.

show() → None

Display all current matplotlib figures.

This method shows all active matplotlib figures using plt.show().

close_plot() → None

Close the current figure.

This method closes the figure stored in self.figure to free up memory.

get_canvas(rows: int = 1, cols: int = 1, shape: tuple[int, int] = (10, 10)) → tuple[Figure, Axes]

Create and return an empty matplotlib canvas for plotting.

Parameters:

• rows – Number of subplot rows (default: 1)

• cols – Number of subplot columns (default: 1)

• shape – Tuple of (width, height) in inches for the figure size (default: (10, 10))

Returns:

A tuple containing:

• matplotlib.figure.Figure: The created figure

• matplotlib.axes.Axes or array of Axes: The created subplot(s)

Return type:

tuple

local_save(path: str)

Save the figure to a local path

Parameters:

path – str:

log(name: str, step: int = None)

Log the current figure using the attached logger.

Parameters:

• name – Name of the logged figure

• step – Optional step or iteration number for the logged figure

Raises:

AssertionError – If no logger is attached to the visualizer

log_image(name: str, step: int = None)

Log the current figure as an image using the attached logger.

This method is an alias for the log() method.

Parameters:

• name – Name of the logged image

• step – Optional step or iteration number for the logged image

Raises:

AssertionError – If no logger is attached to the visualizer

log_table(table: dict, name: str, step: int = None)

Log a dictionary as a table using the attached logger.

Parameters:

• table – Dictionary containing the table data

• name – Name of the logged table

• step – Optional step or iteration number for the logged table

Raises:

AssertionError – If no logger is attached to the visualizer

detoxai.visualization.enums submodule

class detoxai.visualization.enums.ConditionOn(value)

Bases: Enum

PROPER_LABEL = 'proper_label'

IMPROPER_LABEL = 'improper_label'

Module contents