Instantly share code, notes, and snippets.

What would you like to do?
A naive Kohonen-Grossberg Counterpropogation Network in Python
__author__ = 'Juxhin Dyrmishi Brigjaj'
import sys
import math
import random
import argparse
def parse_args():
parser = argparse.ArgumentParser(description='A naive Kohonen-Grossberg Counterpropogation Network in Python')
parser.add_argument('-l', '--learning-rate', metavar='R', type=float, required=True,
help='Float indicating the learning rate (step) the network should use')
parser.add_argument('-f', '--csv-file', type=str, required=True,
help='Path to CSV file containing dataset')
parser.add_argument('-e', '--epoch', type=int, help="Number of epochs to complete", required=True, default=1000)
parser.add_argument('-n', '--neurons', type=int, help="Number of neurons (units) to generate", required=True, default=3)
return parser.parse_args()
def normalise(rows: list=()) -> list:
_result = []
for row in rows:
_vector_length = math.sqrt(sum([x**2 for x in row]))
_result.append([round(x / _vector_length, 4) for x in row])
return _result
def generate_random_units(col_len: int, row_len: int) -> list:
_result = []
for _ in range(0, row_len):
_result.append([round(random.uniform(0.0, 1.0), 4) for _ in range(0, col_len)])
return _result
def calculate_nets(row, units):
_nets = []
for unit in units:
_net = 0.0
for i, _ in enumerate(unit):
_net += round(row[i] * unit[i], 4)
_nets.append(round(_net, 4))
return _nets
def update_units(learning_rate: float, nets: list, row: list, units: list) -> bool:
_i = nets.index(max(nets))
for _j, column in enumerate(row):
units[_i][_j] = round(units[_i][_j] + learning_rate * (column - units[_i][_j]), 4)
def main():
args = parse_args()
learning_rate = args.learning_rate
unnormalised_dataset = []
with open(args.csv_file, 'r') as csv_file:
for line in csv_file:
unnormalised_dataset.append([float(x) for x in line.split(',')])
except TypeError as e:
print("[!] FATAL: Dataset is malformed. Unable to parse values as floats.\n{}".format(str(e)))
print("[+] Normalising dataset")
rows = normalise(unnormalised_dataset)
for row in rows:
print('\t'.join([str(x) for x in row]))
# Used to determine the number of columns in generate_random_units call
# assuming that the dataset is consistent in width
__unit_length = len(unnormalised_dataset[0])
random_units = generate_random_units(__unit_length, args.neurons)
print("\n[+] Starting Weights:")
for unit in random_units:
print(','.join([str(x) for x in unit]))
for i in range(1, args.epoch + 1):
if i % 100 == 0:
print("[+] Running Epoch #{}".format(str(i)))
for row in rows:
nets = calculate_nets(row, random_units)
update_units(learning_rate, nets, row, random_units)
print("\n[+] Final Weights:")
for unit in random_units:
print(','.join([str(x) for x in unit]))
if __name__ == '__main__':
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment