dlpy.model.Model.heat_map_analysis

Model.heat_map_analysis(data=None, mask_width=None, mask_height=None, step_size=None, display=True, img_type='A', image_id=None, filename=None, inputs='_image_', target='_label_', max_display=5, **kwargs)

Conduct a heat map analysis on table of images

Parameters:
data : ImageTable, optional

If data is None then the results from model.predict are used. data specifies the table containing the image data which must contain the columns ‘_image_’, ‘_label_’, ‘_id_’ and ‘_filename_0’.

mask_width : int, optional

Specifies the width of the mask which cover the region of the image.

mask_height : int, optional

Specifies the height of the mask which cover the region of the image.

step_size : int, optional

Specifies the step size of the movement of the the mask.

display : bool, optional

Specifies whether to display the results.

img_type : string, optional

Can be ‘A’ for all images, ‘C’ for only correctly classified images, or ‘M’ for misclassified images.

image_id : list or int, optional

A unique image id to get the heatmap. A standard column of ImageTable

filename : list of strings or string, optional

The name of a file in ‘_filename_0’ if not unique returns multiple

inputs : string, optional

Name of image column for the input into the model.predict function

target : string, optional

Name of column for the correct label

max_display : int, optional

Maximum number of images to display. Heatmap takes a significant amount of time to run so a max of 5 is default.

**kwargs : keyword arguments, optional

Specifies the optional arguments for the dlScore action. For more details, see deepLearn.dlScore

Returns:
pandas.DataFrame

Contains Columns: [‘I__label_’, ‘P__label_(for each label)’, ‘_filename_0’,

‘_id_’, ‘_image_’, ‘_label_’, ‘heat_map’]

Notes

Heat map indicates the important region related with classification. Details of the process can be found at: https://arxiv.org/pdf/1311.2901.pdf.