nogilnick / pythonml

Artificial neural network classes and tools in Python and TensorFlow.

pythonml

Repository containing various python machine learning modules. Several of the modules have corresponding blog plosts on my website: https://nicholastsmith.wordpress.com/.

bprop.py

A straightforward impelementation of the backpropagation algorithm for MLP networks. See this blog post for more information.

DeepOCR

An implementation of OCR using TensorFlow. See the related blog post for more details. Sample code (requires a model to be trained):

from skimage.io import imread
#Takes a while to load model
from DeepOCR import ImageToString
A = imread('Foo.png')
S = ImageToString(A)
print(S)

TFANN

A neural network module containing implementations of MLP, and CNN networks in TensorFlow. The classes in the module adhere to the scikit-learn fit, predict, score interface. Install the package using pip.

pip install TFANN

Sample code for building an MLP regression model:

import numpy as np
from TFANN import ANNR

A = np.random.rand(32, 4)
Y = np.random.rand(32, 1)
a = ANNR([4], [('F', 4), ('AF', 'tanh'), ('F', 1)], maxIter = 16, name = 'mlpr1')
a.fit(A, Y)
S = a.score(A, Y)
YH = a.predict(A)

For building an MLP classification model:

import numpy as np
from TFANN import ANNC

A = np.random.rand(32, 4)
Y = np.array((16 * [1]) + (16 * [0]))
a = ANNC([4], [('F', 4), ('AF', 'tanh'), ('F', 2)], maxIter = 16, name = 'mlpc2')
a.fit(A, Y)
S = a.score(A, Y)
YH = a.predict(A)

For building an CNN classification model:

import numpy as np
from TFANN import ANNC

A = np.random.rand(32, 9, 9, 3)
Y = np.array((16 * [1]) + (16 * [0]))
ws = [('C', [3, 3, 3, 4], [1, 1, 1, 1]), ('AF', 'relu'), 
      ('P', [1, 4, 4, 1], [1, 2, 2, 1]), ('F', 16), 
      ('AF', 'relu'), ('F', 2)]
a = ANNC([9, 9, 3], ws, maxIter = 12, name = "cnnc1")
a.fit(A, Y)
S = a.score(A, Y)
YH = a.predict(A)

TheanoANN.py

A neural network module containing implementations of MLP networks in Theano. The classes in the module adhere to the scikit-learn fit, predict, score interface.