/PyWANN

Python Weightless Artificial Neural Network

Primary LanguagePythonGNU General Public License v2.0GPL-2.0

PyWANN

Python Weightless Artificial Neural Network

Instructions

How to Install

pip install git+git://github.com/firmino/PyWANN.git

Inputs

Python lists or Numpy array.

X is a list of retinas and y is a list of class related with each retina.

ex:

 X = [ [0, 1, 0, 0, 0, 0, 0, 0],
       [0, 0, 1, 1, 1, 1, 0, 0],
       [0, 0, 1, 0, 0, 0, 1, 0],
       [1, 0, 0, 0, 0, 0, 0, 1],
       [1, 1, 0, 1, 1, 1, 1, 1],
       [1, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 1, 0, 0, 1],
       [1, 0, 0, 0, 0, 0, 0, 1]]

y = ['class_a','class_a','class_b','class_b','class_a','class_a','class_b','class_a',]

WiSARD Parameters

  1. retina_length: (INT) The length of the retina
  2. num_bits_addr: (INT) Number of bits used to build the memories
  3. bleaching : (BOOLEAN) If bleaching technique is active or not, default value is True
  4. confidence_threshold: (FLOAT) Confidence used by bleaching technique to solve tie problems, default is 0.1
  5. ignore_zero_addr : (BOOLEAN) The classification of sparse feature vectors can be improved excluding positions zeros. Default value is False
  6. defaul_b_bleaching : (INT) The initial value for bleaching technique. Default value is 1
  7. randomize_positions: (BOOLEAN) If the pseudo-random-mapping will be used or not. Default value is True.
  8. memory_is_cumulative: (BOOLEAN) If false memories store 0 or 1, if true memories count the number of occurrences of patterns. Default value is True

Basic WiSARD (Without Bleaching)

  1. Define the number of bits for each memory
  2. Define the retina's length
  3. Set Bleaching to FALSE
  4. Create a Wisard
  5. Fit with examples (trainning)
  6. Predict (classify)
retina_length = 64
num_bits_addr = 2
bleaching = False

w = WiSARD(retina_length, num_bits_addr, bleaching)


# training discriminators
w.fit(X, y)


# predicting class
result = w.predict(X_test)  #  Result will be a dictionary using the classes as key and the WiSARD result as values

Bleaching WiSARD

  1. Define the number of bits for each memory
  2. Define the retina's length
  3. Set Bleaching to FALSE
  4. Create a Wisard
  5. Fit with examples (trainning)
  6. Predict (classify)
retina_length = 64
num_bits_addr = 2

w = WiSARD(retina_length, num_bits_addr)

# training discriminators
w.fit(X, y)

# predicting class
result = w.predict(X_test)  #  Result will be a dictionary using the classes as key and the WiSARD result as values