layered_view#

scikitplot.visualkeras.layered_view(model, to_file=None, min_z=20, min_xy=20, max_z=400, max_xy=2000, scale_z=0.1, scale_xy=4, type_ignore=None, index_ignore=None, color_map=None, one_dim_orientation='z', index_2d=None, background_fill='white', draw_volume=True, draw_reversed=False, padding=10, text_callable=None, text_vspacing=4, spacing=10, draw_funnel=True, shade_step=10, legend=False, legend_text_spacing_offset=15, font=None, font_color='black', show_dimension=False)[source]#

Generates an architectural visualization for a given linear Keras tf.keras.Model model (i.e., one input and output tensor for each layer) in a layered style, which is particularly suitable for convolutional neural networks (CNNs).

Parameters:

modeltensorflow.keras.Model: A Keras tf.keras.Model model to be visualized.
to_filestr or None: Path to the file where the generated image will be saved. If the image does not exist yet it will be created, else overwritten. The file type is inferred from the file extension. If None, no file is created.
min_zint: Minimum z-dimension size (in pixels) for a layer.
min_xyint: Minimum x- and y-dimension size (in pixels) for a layer.
max_zint: Maximum z-dimension size (in pixels) for a layer.
max_xyint: Maximum x- and y-dimension size (in pixels) for a layer.
scale_zfloat: Scalar multiplier for the z-dimension size of each layer.
scale_xyfloat: Scalar multiplier for the x- and y-dimension size of each layer.
type_ignorelist of str: List of layer types to ignore when visualizing the model.
index_ignorelist of int: List of layer indices to ignore when visualizing the model.
color_mapdict: A dictionary mapping layer defining fill and outline for each layer by class type. Layers not specified in the dictionary will use default colors.
one_dim_orientation{‘x’, ‘y’, ‘z’}: Axis along which one-dimensional layers should be drawn.
index_2dlist of int: Indices of layers to be drawn in 2D when draw_volume is True.
background_fillstr or tuple: Background color of the image. Can be a string or a tuple (R, G, B, A).
draw_volumebool: Whether to use a 3D volumetric view (True) or a 2D box view (False).
draw_reversedbool: Whether to draw 3D boxes in reverse order, from front-right to back-left.
paddingint: Distance in pixels before the first and after the last layer.
text_callablecallable: A callable that generates text for layers. The callable should take two arguments: the layer index (int) and the layer (Layer).
text_vspacingint: Vertical spacing in pixels between lines of text produced by text_callable.
spacingint: Horizontal spacing in pixels between consecutive layers.
draw_funnelbool: If set to True, a funnel will be drawn between consecutive layers.
shade_stepfloat: Lightness deviation step for shades in the visualization (only applicable in 3D volumetric view).
legendbool: Whether to include a legend of the layers in the image.
legend_text_spacing_offsetfloat: Offset for the space allocated to legend text. Useful for preventing text cutoff in the legend.
fontstr or None: Font to be used for legend text. If None, the default font is used.
font_colorstr or tuple: Color of the font. Can be a string or a tuple (R, G, B, A).
show_dimensionbool: Whether to display layer dimensions in the legend (only when legend is True).

Returns:

image: The generated architecture visualization image.

Parameters:

to_file (str)
min_z (int)
min_xy (int)
max_z (int)
max_xy (int)
scale_z (float)
scale_xy (float)
type_ignore (list)
index_ignore (list)
color_map (dict)
one_dim_orientation (str)
index_2d (list)
background_fill (Any)
draw_volume (bool)
draw_reversed (bool)
padding (int)
text_callable (Callable[[int, Layer], tuple])
text_vspacing (int)
spacing (int)
draw_funnel (bool)
legend (bool)
font (<module 'PIL.ImageFont' from '/opt/conda/lib/python3.11/site-packages/PIL/ImageFont.py'>)
font_color (Any)

Return type: