White-Box-Layer is a Python module for deep learning built on top of TensorFlow and is distributed under the MIT license.
The project was started in May 2021 by YeongHyeon Park.
This project does not limit for participation.
Contribute now!
Installation
Dependencies
whiteboxlayer requires:
- Numpy: 1.18.5
- Scipy: 1.4.1
- TensorFlow: 2.3.0
User installation
You can install the white-box-layer via simple command as below.
$ pip install whiteboxlayer
Development
We welcome new contributors of all experience levels. The white-box-layer community goals are to be helpful, welcoming, and effective. The Development Guide has detailed information about contributing code, documentation, tests, and more. We've included some basic information in this README.
Example
Example for Convolutional Neural Network
An example of constructing a convolutional neural network is covered. The relevant source code is additionally provided following links.
Define TensorFlow based module
class Neuralnet(tf.Module):
def __init__(self, **kwargs):
super(Neuralnet, self).__init__()
self.who_am_i = kwargs['who_am_i']
self.dim_h = kwargs['dim_h']
self.dim_w = kwargs['dim_w']
self.dim_c = kwargs['dim_c']
self.num_class = kwargs['num_class']
self.filters = kwargs['filters']
self.layer = wbl.Layers()
self.forward = tf.function(self.__call__)
@tf.function
def __call__(self, x, verbose=False):
logit = self.__nn(x=x, name=self.who_am_i, verbose=verbose)
y_hat = tf.nn.softmax(logit, name="y_hat")
return logit, y_hat
def __nn(self, x, name='neuralnet', verbose=True):
for idx, _ in enumerate(self.filters[:-1]):
if(idx == 0): continue
x = self.layer.conv2d(x=x, stride=1, \
filter_size=[3, 3, self.filters[idx-1], self.filters[idx]], \
activation='relu', name='%s-%dconv' %(name, idx), verbose=verbose)
x = self.layer.maxpool(x=x, ksize=2, strides=2, \
name='%s-%dmp' %(name, idx), verbose=verbose)
x = tf.reshape(x, shape=[x.shape[0], -1], name="flat")
x = self.layer.fully_connected(x=x, c_out=self.filters[-1], \
activation='relu', name="%s-clf0" %(name), verbose=verbose)
x = self.layer.fully_connected(x=x, c_out=self.num_class, \
activation=None, name="%s-clf1" %(name), verbose=verbose)
return x
Initializing module
model = Neuralnet(\
who_am_i="CNN", \
dim_h=28, dim_w=28, dim_c=1, \
num_class=10, \
filters=[1, 32, 64, 128])
dummy = tf.zeros((1, model.dim_h, model.dim_w, model.dim_c), dtype=tf.float32)
model.forward(x=dummy, verbose=True)
Results
Conv (CNN-1conv) (1, 28, 28, 1) -> (1, 28, 28, 32)
MaxPool (CNN-1mp) (1, 28, 28, 32) > (1, 14, 14, 32)
Conv (CNN-2conv) (1, 14, 14, 32) -> (1, 14, 14, 64)
MaxPool (CNN-2mp) (1, 14, 14, 64) > (1, 7, 7, 64)
FC (CNN-clf0) (1, 3136) -> (1, 128)
FC (CNN-clf1) (1, 128) -> (1, 10)
Conv (CNN-1conv) (1, 28, 28, 1) -> (1, 28, 28, 32)
MaxPool (CNN-1mp) (1, 28, 28, 32) > (1, 14, 14, 32)
Conv (CNN-2conv) (1, 14, 14, 32) -> (1, 14, 14, 64)
MaxPool (CNN-2mp) (1, 14, 14, 64) > (1, 7, 7, 64)
FC (CNN-clf0) (1, 3136) -> (1, 128)
FC (CNN-clf1) (1, 128) -> (1, 10)
2 Aug 21, 2022
9.6k Jan 2, 2023
6.2k Jan 1, 2023
1.8k Dec 28, 2022
437 Jan 3, 2023
127 Dec 12, 2022
5.7k Jan 9, 2023
3.4k Jan 7, 2023
84 Nov 23, 2022
4 Dec 01, 2022
36 Dec 12, 2022
171 Dec 29, 2022
52 Dec 22, 2022
15 Dec 22, 2022
1 Dec 16, 2021
24 Oct 26, 2022
7 Dec 15, 2022
631 Jan 04, 2023
58 Dec 17, 2022
24 Nov 24, 2022
42 Nov 24, 2022
2 Oct 25, 2021
5 May 28, 2021
4k Dec 30, 2022
40 Nov 20, 2022
15 Dec 22, 2022
7 Apr 13, 2022
87 Jan 07, 2023