Contents

• ikkuna.visualization
  • Module contents

ikkuna.visualization

Module contents

class ikkuna.visualization.TBBackend(**kwargs)

Bases: ikkuna.visualization.backend.Backend

Tensorboard backend.

Note

Whitespace and punctuation in the title will be replaced with underscores due to the fact that it becomes part of a file name.

_writer

Type:tensorboardX.SummaryWriter

_hist_bins

Number of bins to use for histograms

Type:int

info = 'n/a'

add_data(module_name, datum, step)

Display scalar data (i.e. a line plot)

Parameters:

• module_name (str) – Name of module which emitted the data
• datum (torch.Tensor) – Payload
• step (int) – Global step

add_histogram(module_name, datum, step, bins='auto')

Display histogram data (i.e. a line plot)

Parameters:

• module_name (str) – Name of module which emitted the data
• datum (torch.Tensor) – Payload, not the histogram itself
• step (int) – Global step

class ikkuna.visualization.MPLBackend(**kwargs)

Bases: ikkuna.visualization.backend.Backend

Matplotlib backend (use in Jupyter with %matplotlib inline or via X-forwarding over ssh [barely useable])

_xlabel

X axis label for line plots

Type:str

_ylabel

Y axis label for line plots

Type:str

_ylims

Limits of the y axis for line plots

Type:tuple

_redraw_counter

Number of datapoints to consume before redrawing the figure

Type:int

_plots

module-plot mapping

Type:dict

_axes

module-UpdatableHistogram mapping (this should be refactored)

Type:dict

_buffer

Per-module buffer of tensors for more reliable histograms

Type:dict

_buffer_lim

Size of the buffer

Type:int

__init__(**kwargs)

Parameters:

• xlabel (str) –
• ylabel (str) –
• ylims (tuple) –
• buffer_lim (int) – Buffer size for more reliable histograms

_prepare_axis(ax)

Prepare the line plot axis with labels and scaling.

title

xlabel

ylabel

_reflow_plots()

Reqorganize the histogram subplots into a rectangular shape. For now, the sublots are arranged on a grid twice as high as it is wide, since vertical space is often in abundance.

Note

This does not recompute the plots

add_histogram(module_name, datum, step)

Display histogram data (i.e. a line plot)

Parameters:

• module_name (str) – Name of module which emitted the data
• datum (torch.Tensor) – Payload, not the histogram itself
• step (int) – Global step

add_data(module_name, datum, step)

Display scalar data (i.e. a line plot)

Parameters:

• module_name (str) – Name of module which emitted the data
• datum (torch.Tensor) – Payload
• step (int) – Global step

class ikkuna.visualization.Backend(title)

Bases: abc.ABC

Base class for visualiation backends. Subscribers use this class to dispatch their metrics to have them visualised.

title

The figure title

Type:str

__init__(title)

Parameters:title (str) – Title to use for the figure.

add_data(module_name, datum, step)

Display scalar data (i.e. a line plot)

Parameters:

• module_name (str) – Name of module which emitted the data
• datum (torch.Tensor) – Payload
• step (int) – Global step

add_histogram(module_name, datum, step)

Display histogram data (i.e. a line plot)

Parameters:

• module_name (str) – Name of module which emitted the data
• datum (torch.Tensor) – Payload, not the histogram itself
• step (int) – Global step

title

ikkuna.visualization.configure_prefix(p)

Set a prefix to the log directory for Tensorboard.

Parameters:p (str) – Prefix to the directory name. _runs will be appended by TBBackend

class ikkuna.visualization.NullBackend(*args, **kwargs)

Bases: ikkuna.visualization.backend.Backend

add_data(*args, **kwargs)

Display scalar data (i.e. a line plot)

Parameters:

• module_name (str) – Name of module which emitted the data
• datum (torch.Tensor) – Payload
• step (int) – Global step

add_histogram(*args, **kwargs)

Display histogram data (i.e. a line plot)

Parameters:

• module_name (str) – Name of module which emitted the data
• datum (torch.Tensor) – Payload, not the histogram itself
• step (int) – Global step

title()