match_nn.py

# coding: utf-8
#
# Simulation program for unplugged neural network model.
# Reference:
# https://www.1101.com/morikawa/2001-03-12.html
# https://www.1101.com/morikawa/2001-04-02.html

import itertools
import random
import numpy as np
random.seed(1)
np.random.seed(1)

prices = np.array([310, 220, 70])
max_choice = 3
max_price = 600
data = [np.array(i) for i in itertools.product(range(max_choice), repeat=len(prices))][::-1]
random.shuffle(data)
train_data = data[:5]
test_data = data[5:]
iters = 3

nodes = np.array([5, 5, 5])
thres = 6
learn_rate = iters

for choice in test_data:
  print 'CHOICE: %s,\tSUM_PRICE: %.2f,\tSUM_WEIGHT:%d\tTHRES:%d\tNodes:%s' % (
      choice, np.dot(prices, choice), np.dot(nodes, choice), thres, nodes)
  if np.dot(nodes, choice) > thres:
    print 'FIRE',
    if np.dot(prices, choice) > max_price:
      print 'ok'
    else:
      print 'ng. too sensitive.'
  else:
    print 'STAY',
    if np.dot(prices, choice) > max_price:
      print 'ng. too insensitive.'
    else:
      print 'ok'
print '>>> TRAINING START <<<'
for iteration in range(iters):
  for choice in train_data:
    print 'CHOICE: %s,\tSUM_PRICE: %.2f,\tSUM_WEIGHT:%d\tTHRES:%d\tNodes:%s' % (
        choice, np.dot(prices, choice), np.dot(nodes, choice), thres, nodes)
    if np.dot(nodes, choice) > thres:
      print 'FIRE',
      if np.dot(prices, choice) > max_price:
        print 'ok'
      else:
        print 'ng. too sensitive.'
        thres += learn_rate
        nodes[choice > 0] -= learn_rate
    else:
      print 'STAY',
      if np.dot(prices, choice) > max_price:
        print 'ng. too insensitive.'
        thres -= learn_rate
        nodes[choice > 0] += learn_rate
      else:
        print 'ok'
    nodes = np.clip(nodes, 0, 10)
  learn_rate -= 1

print '>>> TRAINING DONE <<<'

for choice in test_data:
  print 'CHOICE: %s,\tSUM_PRICE: %.2f,\tSUM_WEIGHT:%d\tTHRES:%d\tNodes:%s' % (
      choice, np.dot(prices, choice), np.dot(nodes, choice), thres, nodes)
  if np.dot(nodes, choice) > thres:
    print 'FIRE',
    if np.dot(prices, choice) > max_price:
      print 'ok'
    else:
      print 'ng. too sensitive.'
  else:
    print 'STAY',
    if np.dot(prices, choice) > max_price:
      print 'ng. too insensitive.'
    else:
      print 'ok'