Litten

Litten is a python package to visualize sequential Tensorflow(keras) neural network model architectures, Get layers summary in details, visualize conv layers filters and featurmaps.

Installation

To install latest version from PyPi

$ pip install litten

How to use it?

1. Create your neural network archtictures

model = keras.models.Sequential(
    [
        keras.layers.Conv2D(filters = 32, kernel_size = 3, input_shape = [150, 150, 3]),
        ...
        keras.layers.Dense(units = 2, activation = 'softmax')
    ]
)

# or using functional API

input = keras.Input(shape=(28, 28, 1), name="img")
...
output = keras.layers.Dense(units = 2, activation = 'softmax')(x)
model = keras.Model(input, output, name="Model")

2. Import LayersSummary, ModelVisualizer and create objects

from litten import LayersSummary, ModelVisualizer

3. To get Layers Summaries

summary = LayersSummary()
summary.show_layers_summaries(model)

Output:

=================================================================================================================
Layer 1: InputLayer     | Attributes
----------------------------------------
built                   :  True
sparse                  :  False
ragged                  :  False
batch_size              :  None
is_placeholder          :  True
=================================================================================================================
Layer 2: Conv2D         | Attributes
----------------------------------------
rank                    :  2
filters                 :  16
groups                  :  1
kernel_size             :  (3, 3)
strides                 :  (1, 1)
padding                 :  valid
data_format             :  channels_last
dilation_rate           :  (1, 1)
activation              :  <function relu at 0x7f6b7038fee0>
use_bias                :  True
kernel_initializer      :  <keras.initializers.initializers_v2.GlorotUniform object at 0x7f6aee189610>
bias_initializer        :  <keras.initializers.initializers_v2.Zeros object at 0x7f6aee189af0>
kernel_regularizer      :  None
bias_regularizer        :  None
kernel_constraint       :  None
bias_constraint         :  None
built                   :  True
=================================================================================================================
....

4. To visualize model architecture

   ModelVisualizer.visualize_model(
       show_names=False,
       show_properties=False,
       show_connectors=False,
       pallete='default'
   )
   

   Example 1

   vis = ModelVisualizer(model)
   vis.visualize_model()

   

   Example 2

   vis.visualize_model(show_names=True)

   

   Example 3

   vis.visualize_model(show_names=True, show_connectors = True)

   

   Exampel 4

   vis.visualize_model(show_names=True, show_connectors = True, show_properties=True)

   

   Example 5 You can choose one from these palettes default, red, green, blue, yellow, brown, purple, gray

   vis.visualize_model(show_names=True, show_connectors=True, palette=<palette>)

   

5. To visualize Conv filters

       ModelVisualizer.visualize_filters(
           cmap = 'gray'   # matplotlib cmaps
       )
   

   Example

   vis = ModelVisualizer(model)
   vis.visualize_filters('Blues')

   

6. To visualize Features map

   ModelVisualizer.visualize_featuremap(
       input_image,
       cmap = 'gray'   # matplotlib cmaps
   )
   

   Example

   vis = ModelVisualizer(model)
   vis.visualize_featuremap(input_image)

   

Contributing to litten

To contribute to litten, follow these steps:

1. Fork this repository.
2. Create a branch: git checkout -b <branch_name>.
3. Make your changes and commit them: git commit -m '<commit_message>'
4. Push to the original branch: git push origin <project_name>/<location>
5. Create the pull request.

Alternatively see the GitHub documentation on creating a pull request.

License

Open source licensed under the MIT license (see LICENSE file for details).

Tools

• Python
• PIL
• Matplotlib
• pytest