visualizer - visualizer class

The file yann.modules.visualizer.py contains the definition for the visualizer:

yann.modules.visualizer.save_images(imgs, prefix, is_color, verbose=2)

This functions produces a visualiation of the filters.

Parameters:

• imgs – images of shape .. [num_imgs, height, width, channels] Note the change in order. it can also be in lots of other shapes and they will be reshaped and saved as images.
• prefix – address to save the image to
• is_color – If the image is color or not. True only if image shape is of color images.
• verbose – As Always

class yann.modules.visualizer.visualizer(visualizer_init_args, verbose=2)

Visualizer saves down images to visualize. The initilizer only initializes the directories for storing visuals. Three types of visualizations are saved down:

• filters of each layer
• activations of each layer
• raw images to check the activations against

Parameters:

• verbose – Similar to any 3-level verbose in the toolbox.
• visualizer_init_args –

  visualer_params is a dictionary of the form:

  visualizer_init_args = {
      "root"       : <location to save the visualizations at>,
      "frequency"  : <integer>, after how many epochs do you need to
                      visualize. Default value is 1
      "sample_size": <integer, prefer squares>, simply save down random
                      images from the datasets saves down activations for the
                      same images also. Default value is 16
      "rgb_filters": <bool> flag. if True a 3D-RGB rendition of the CNN
                      filters is rendered. Default value is False.
      "debug_functions" : <bool> visualize train and test and other theano functions.
                          default is False. Needs pydot and dv2viz to be installed.
      "debug_layers" : <bool> Will print layer activities from input to that layer
                       output. ( this is almost always useless because test debug
                       function will combine all these layers and print directly.)
      "id"         : id of the visualizer
                  }
  

Returns:

A visualizer object.

Return type:

yann.modules.visualizer

initialize(batch_size, verbose=2)

Function that cooks the visualizer for some dataset.

Parameters:

• batch_size – form dataset
• verbose – as always

theano_function_visualizer(function, short_variable_names=False, format='pdf', verbose=2)

This basically prints a visualization of any theano function using the in-built theano visualizer. It will save both a interactive html file and a plain old png file. This is just a wrapper to theano’s visualization tools.

Parameters:

• function – theano function to print
• short_variable_names – If True will print variables in short.
• format – Any pydot supported format. Default is ‘pdf’
• verbose – As usual.

visualize_activities(layer_activities, epoch, index=0, verbose=2)

This method saves down all activities.

Parameters:

• layer_activities – network’s layer_activities as created
• epoch – what epoch are we running currently.
• verbose – as always

visualize_filters(layers, epoch, index=0, verbose=2)

This method saves down all activities.

Parameters:

• layers – network’s layer dictionary
• epoch – what epoch are we running currently.
• verbose – as always

visualize_images(imgs, loc=None, verbose=2)

Visualize the images in the dataset. Assumes that the data in the tensor variable imgs is in shape (batch_size, height, width, channels). Assumes that batchsize does not change.

Parameters:

• imgs – tensor of data
• verbose – as usual.