API tools faq
paste
Advertisement
cimona

Normalizacija

cimona
Sep 14th, 2018
341
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 11.97 KB | None | 0 0
raw download clone embed print report
  1. from __future__ import division
  2. from math import exp
  3. from random import seed
  4. from random import random
  5. from collections import defaultdict
  6.  
  7. # Initialize a network
  8. # Put fixed weigths to 0.5 on code.finki.ukim.mk if there is a problem with random()
  9. def initialize_network(n_inputs, n_hidden, n_outputs):
  10. network = list()
  11. hidden_layer = [{'weights':[random() for i in range(n_inputs + 1)]} for i in range(n_hidden)]
  12. network.append(hidden_layer)
  13. output_layer = [{'weights':[random() for i in range(n_hidden + 1)]} for i in range(n_outputs)]
  14. network.append(output_layer)
  15. return network
  16.  
  17.  
  18. # Calculate neuron activation for an input
  19. def activate(weights, inputs):
  20. activation = weights[-1]
  21. for i in range(len(weights)-1):
  22. activation += weights[i] * inputs[i]
  23. return activation
  24.  
  25. # Transfer neuron activation
  26. def transfer(activation):
  27. return 1.0 / (1.0 + exp(-activation))
  28. def predict(network, row):
  29. outputs = forward_propagate(network,row)
  30. return outputs.index(max(outputs))
  31. # Forward propagate input to a network output
  32. def forward_propagate(network, row):
  33. inputs = row
  34. for layer in network:
  35. new_inputs = []
  36. for neuron in layer:
  37. activation = activate(neuron['weights'], inputs)
  38. neuron['output'] = transfer(activation)
  39. new_inputs.append(neuron['output'])
  40. inputs = new_inputs
  41. return inputs
  42.  
  43. # Calculate the derivative of an neuron output
  44. def transfer_derivative(output):
  45. return output * (1.0 - output)
  46.  
  47. # Backpropagate error and store in neurons
  48. def backward_propagate_error(network, expected):
  49. for i in reversed(range(len(network))):
  50. layer = network[i]
  51. errors = list()
  52. if i != len(network)-1:
  53. for j in range(len(layer)):
  54. error = 0.0
  55. for neuron in network[i + 1]:
  56. error += (neuron['weights'][j] * neuron['delta'])
  57. errors.append(error)
  58. else:
  59. for j in range(len(layer)):
  60. neuron = layer[j]
  61. errors.append(expected[j] - neuron['output'])
  62. for j in range(len(layer)):
  63. neuron = layer[j]
  64. neuron['delta'] = errors[j] * transfer_derivative(neuron['output'])
  65.  
  66. # Update network weights with error
  67. def update_weights(network, row, l_rate):
  68. for i in range(len(network)):
  69. inputs = row[:-1]
  70. if i != 0:
  71. inputs = [neuron['output'] for neuron in network[i - 1]]
  72. for neuron in network[i]:
  73. for j in range(len(inputs)):
  74. neuron['weights'][j] += l_rate * neuron['delta'] * inputs[j]
  75. neuron['weights'][-1] += l_rate * neuron['delta']
  76.  
  77. # Train a network for a fixed number of epochs
  78. def train_network(network, train, l_rate, n_epoch, n_outputs):
  79. for epoch in range(n_epoch):
  80. sum_error = 0
  81. for row in train:
  82. outputs = forward_propagate(network, row)
  83. expected = [0 for i in range(n_outputs)]
  84. expected[int(row[-1])] = 1
  85. sum_error += sum([(expected[i]-outputs[i])**2 for i in range(len(expected))])
  86. backward_propagate_error(network, expected)
  87. update_weights(network, row, l_rate)
  88.  
  89.  
  90. # Test training backprop algorithm
  91. seed(1)
  92. dataset = [[2.7810836,2.550537003,0],
  93. [1.465489372,2.362125076,0],
  94. [3.396561688,4.400293529,0],
  95. [1.38807019,1.850220317,0],
  96. [3.06407232,3.005305973,0],
  97. [7.627531214,2.759262235,1],
  98. [5.332441248,2.088626775,1],
  99. [6.922596716,1.77106367,1],
  100. [8.675418651,-0.242068655,1],
  101. [7.673756466,3.508563011,1]]
  102.  
  103. n_inputs = len(dataset[0]) - 1
  104. n_outputs = len(set([row[-1] for row in dataset]))
  105. network = initialize_network(n_inputs, 2, n_outputs)
  106. train_network(network, dataset, 0.5, 20, n_outputs)
  107.  
  108.  
  109. seed(1)
  110. training_data = [
  111. [3.6216, 8.6661, -2.8073, -0.44699, 0],
  112. [4.5459, 8.1674, -2.4586, -1.4621, 0],
  113. [3.866, -2.6383, 1.9242, 0.10645, 0],
  114. [3.4566, 9.5228, -4.0112, -3.5944, 0],
  115. [0.32924, -4.4552, 4.5718, -0.9888, 0],
  116. [4.3684, 9.6718, -3.9606, -3.1625, 0],
  117. [3.5912, 3.0129, 0.72888, 0.56421, 0],
  118. [2.0922, -6.81, 8.4636, -0.60216, 0],
  119. [3.2032, 5.7588, -0.75345, -0.61251, 0],
  120. [1.5356, 9.1772, -2.2718, -0.73535, 0],
  121. [1.2247, 8.7779, -2.2135, -0.80647, 0],
  122. [3.9899, -2.7066, 2.3946, 0.86291, 0],
  123. [1.8993, 7.6625, 0.15394, -3.1108, 0],
  124. [-1.5768, 10.843, 2.5462, -2.9362, 0],
  125. [3.404, 8.7261, -2.9915, -0.57242, 0],
  126. [4.6765, -3.3895, 3.4896, 1.4771, 0],
  127. [2.6719, 3.0646, 0.37158, 0.58619, 0],
  128. [0.80355, 2.8473, 4.3439, 0.6017, 0],
  129. [1.4479, -4.8794, 8.3428, -2.1086, 0],
  130. [5.2423, 11.0272, -4.353, -4.1013, 0],
  131. [5.7867, 7.8902, -2.6196, -0.48708, 0],
  132. [0.3292, -4.4552, 4.5718, -0.9888, 0],
  133. [3.9362, 10.1622, -3.8235, -4.0172, 0],
  134. [0.93584, 8.8855, -1.6831, -1.6599, 0],
  135. [4.4338, 9.887, -4.6795, -3.7483, 0],
  136. [0.7057, -5.4981, 8.3368, -2.8715, 0],
  137. [1.1432, -3.7413, 5.5777, -0.63578, 0],
  138. [-0.38214, 8.3909, 2.1624, -3.7405, 0],
  139. [6.5633, 9.8187, -4.4113, -3.2258, 0],
  140. [4.8906, -3.3584, 3.4202, 1.0905, 0],
  141. [-0.24811, -0.17797, 4.9068, 0.15429, 0],
  142. [1.4884, 3.6274, 3.308, 0.48921, 0],
  143. [4.2969, 7.617, -2.3874, -0.96164, 0],
  144. [-0.96511, 9.4111, 1.7305, -4.8629, 0],
  145. [-1.6162, 0.80908, 8.1628, 0.60817, 0],
  146. [2.4391, 6.4417, -0.80743, -0.69139, 0],
  147. [2.6881, 6.0195, -0.46641, -0.69268, 0],
  148. [3.6289, 0.81322, 1.6277, 0.77627, 0],
  149. [4.5679, 3.1929, -2.1055, 0.29653, 0],
  150. [3.4805, 9.7008, -3.7541, -3.4379, 0],
  151. [4.1711, 8.722, -3.0224, -0.59699, 0],
  152. [-0.2062, 9.2207, -3.7044, -6.8103, 0],
  153. [-0.0068919, 9.2931, -0.41243, -1.9638, 0],
  154. [0.96441, 5.8395, 2.3235, 0.066365, 0],
  155. [2.8561, 6.9176, -0.79372, 0.48403, 0],
  156. [-0.7869, 9.5663, -3.7867, -7.5034, 0],
  157. [2.0843, 6.6258, 0.48382, -2.2134, 0],
  158. [-0.7869, 9.5663, -3.7867, -7.5034, 0],
  159. [3.9102, 6.065, -2.4534, -0.68234, 0],
  160. [1.6349, 3.286, 2.8753, 0.087054, 0],
  161. [4.3239, -4.8835, 3.4356, -0.5776, 0],
  162. [5.262, 3.9834, -1.5572, 1.0103, 0],
  163. [3.1452, 5.825, -0.51439, -1.4944, 0],
  164. [2.549, 6.1499, -1.1605, -1.2371, 0],
  165. [4.9264, 5.496, -2.4774, -0.50648, 0],
  166. [4.8265, 0.80287, 1.6371, 1.1875, 0],
  167. [2.5635, 6.7769, -0.61979, 0.38576, 0],
  168. [5.807, 5.0097, -2.2384, 0.43878, 0],
  169. [3.1377, -4.1096, 4.5701, 0.98963, 0],
  170. [-0.78289, 11.3603, -0.37644, -7.0495, 0],
  171. [-1.3971, 3.3191, -1.3927, -1.9948, 1],
  172. [0.39012, -0.14279, -0.031994, 0.35084, 1],
  173. [-1.6677, -7.1535, 7.8929, 0.96765, 1],
  174. [-3.8483, -12.8047, 15.6824, -1.281, 1],
  175. [-3.5681, -8.213, 10.083, 0.96765, 1],
  176. [-2.2804, -0.30626, 1.3347, 1.3763, 1],
  177. [-1.7582, 2.7397, -2.5323, -2.234, 1],
  178. [-0.89409, 3.1991, -1.8219, -2.9452, 1],
  179. [0.3434, 0.12415, -0.28733, 0.14654, 1],
  180. [-0.9854, -6.661, 5.8245, 0.5461, 1],
  181. [-2.4115, -9.1359, 9.3444, -0.65259, 1],
  182. [-1.5252, -6.2534, 5.3524, 0.59912, 1],
  183. [-0.61442, -0.091058, -0.31818, 0.50214, 1],
  184. [-0.36506, 2.8928, -3.6461, -3.0603, 1],
  185. [-5.9034, 6.5679, 0.67661, -6.6797, 1],
  186. [-1.8215, 2.7521, -0.72261, -2.353, 1],
  187. [-0.77461, -1.8768, 2.4023, 1.1319, 1],
  188. [-1.8187, -9.0366, 9.0162, -0.12243, 1],
  189. [-3.5801, -12.9309, 13.1779, -2.5677, 1],
  190. [-1.8219, -6.8824, 5.4681, 0.057313, 1],
  191. [-0.3481, -0.38696, -0.47841, 0.62627, 1],
  192. [0.47368, 3.3605, -4.5064, -4.0431, 1],
  193. [-3.4083, 4.8587, -0.76888, -4.8668, 1],
  194. [-1.6662, -0.30005, 1.4238, 0.024986, 1],
  195. [-2.0962, -7.1059, 6.6188, -0.33708, 1],
  196. [-2.6685, -10.4519, 9.1139, -1.7323, 1],
  197. [-0.47465, -4.3496, 1.9901, 0.7517, 1],
  198. [1.0552, 1.1857, -2.6411, 0.11033, 1],
  199. [1.1644, 3.8095, -4.9408, -4.0909, 1],
  200. [-4.4779, 7.3708, -0.31218, -6.7754, 1],
  201. [-2.7338, 0.45523, 2.4391, 0.21766, 1],
  202. [-2.286, -5.4484, 5.8039, 0.88231, 1],
  203. [-1.6244, -6.3444, 4.6575, 0.16981, 1],
  204. [0.50813, 0.47799, -1.9804, 0.57714, 1],
  205. [1.6408, 4.2503, -4.9023, -2.6621, 1],
  206. [0.81583, 4.84, -5.2613, -6.0823, 1],
  207. [-5.4901, 9.1048, -0.38758, -5.9763, 1],
  208. [-3.2238, 2.7935, 0.32274, -0.86078, 1],
  209. [-2.0631, -1.5147, 1.219, 0.44524, 1],
  210. [-0.91318, -2.0113, -0.19565, 0.066365, 1],
  211. [0.6005, 1.9327, -3.2888, -0.32415, 1],
  212. [0.91315, 3.3377, -4.0557, -1.6741, 1],
  213. [-0.28015, 3.0729, -3.3857, -2.9155, 1],
  214. [-3.6085, 3.3253, -0.51954, -3.5737, 1],
  215. [-6.2003, 8.6806, 0.0091344, -3.703, 1],
  216. [-4.2932, 3.3419, 0.77258, -0.99785, 1],
  217. [-3.0265, -0.062088, 0.68604, -0.055186, 1],
  218. [-1.7015, -0.010356, -0.99337, -0.53104, 1],
  219. [-0.64326, 2.4748, -2.9452, -1.0276, 1],
  220. [-0.86339, 1.9348, -2.3729, -1.0897, 1],
  221. [-2.0659, 1.0512, -0.46298, -1.0974, 1],
  222. [-2.1333, 1.5685, -0.084261, -1.7453, 1],
  223. [-1.2568, -1.4733, 2.8718, 0.44653, 1],
  224. [-3.1128, -6.841, 10.7402, -1.0172, 1],
  225. [-4.8554, -5.9037, 10.9818, -0.82199, 1],
  226. [-2.588, 3.8654, -0.3336, -1.2797, 1],
  227. [0.24394, 1.4733, -1.4192, -0.58535, 1],
  228. [-1.5322, -5.0966, 6.6779, 0.17498, 1],
  229. [-4.0025, -13.4979, 17.6772, -3.3202, 1],
  230. [-4.0173, -8.3123, 12.4547, -1.4375, 1]
  231. ]
  232.  
  233. testing_data = [
  234. [2.888, 0.44696, 4.5907, -0.24398, 0],
  235. [0.49665, 5.527, 1.7785, -0.47156, 0],
  236. [4.2586, 11.2962, -4.0943, -4.3457, 0],
  237. [1.7939, -1.1174, 1.5454, -0.26079, 0],
  238. [5.4021, 3.1039, -1.1536, 1.5651, 0],
  239. [2.5367, 2.599, 2.0938, 0.20085, 0],
  240. [4.6054, -4.0765, 2.7587, 0.31981, 0],
  241. [-1.979, 3.2301, -1.3575, -2.5819, 1],
  242. [-0.4294, -0.14693, 0.044265, -0.15605, 1],
  243. [-2.234, -7.0314, 7.4936, 0.61334, 1],
  244. [-4.211, -12.4736, 14.9704, -1.3884, 1],
  245. [-3.8073, -8.0971, 10.1772, 0.65084, 1],
  246. [-2.5912, -0.10554, 1.2798, 1.0414, 1],
  247. [-2.2482, 3.0915, -2.3969, -2.6711, 1]
  248. ]
  249.  
  250. def findMax(data):
  251. maximums = []
  252. for i in range(0,len(data[0])-1):
  253. MAX = 0
  254. for j in range(0,len(data)):
  255. if data[j][i] > MAX:
  256. MAX = data[j][i]
  257. maximums.append(MAX)
  258.  
  259. return maximums
  260.  
  261. def findMin(data):
  262. minimums = []
  263. for i in range(0, len(data[0]) - 1):
  264. MIN = 5000
  265. for j in range(0, len(data)):
  266. if data[j][i] < MIN:
  267. MIN = data[j][i]
  268.  
  269. minimums.append(MIN)
  270.  
  271. return minimums
  272.  
  273. # f-ja za normalizacija
  274. def minMaxNrom(training_data, testing_data):
  275. MAX=0
  276. MIN=0
  277. new_data =[]
  278. new_testing_data=[]
  279.  
  280. # trainingData normalization
  281. for i in range(0, len(training_data)):
  282. row = []
  283.  
  284. for j in range(0, len(training_data[i])-1):
  285. MIN = findMin(training_data)
  286. MAX = findMax(training_data)
  287. row.append((training_data[i][j] - MIN[j])/(MAX[j] - MIN[j]))
  288.  
  289. row.append(training_data[i][-1])
  290. new_data.append(row)
  291.  
  292. #testingData normalization
  293. for i in range(0, len(testing_data)):
  294. row = []
  295.  
  296. for j in range(0, len(testing_data[i])-1):
  297. MIN = findMin(testing_data)
  298. MAX = findMax(testing_data)
  299. row.append((testing_data[i][j] - MIN[j]) / (MAX[j] - MIN[j]))
  300.  
  301. row.append(testing_data[i][-1])
  302. new_testing_data.append(row)
  303.  
  304. return (new_data, new_testing_data)
  305.  
  306.  
  307. if __name__ == "__main__":
  308. #unnormalized data
  309. n_inputs=len(training_data[0])-1
  310. n_outputs=len(set(row[-1] for row in training_data))
  311. network = initialize_network(n_inputs,3,n_outputs)
  312. train_network(network,training_data,0.3,50,n_outputs)
  313.  
  314. #normalized data
  315. Ntraining = minMaxNrom(training_data,testing_data)[0]
  316. Ntesting = minMaxNrom(training_data,testing_data)[1]
  317. n_inputs=len(Ntraining[0])-1
  318. n_outputs=len(set(row[-1] for row in Ntraining))
  319. Nnetwork = initialize_network(n_inputs,3,n_outputs)
  320. train_network(Nnetwork,Ntraining,0.3,50,n_outputs)
  321.  
  322.  
  323. correct_first = 0
  324. correct_second = 0
  325.  
  326. #testiraj nenormalizirana mreza
  327. for row in testing_data:
  328. prediction = predict(network, row)
  329. if (row[-1] == prediction):
  330. correct_first += 1
  331.  
  332.  
  333. #testiraj normalizirana mreza
  334. for row in Ntesting:
  335. prediction = predict(Nnetwork, row)
  336. if (row[-1] == prediction):
  337. correct_second += 1
  338.  
  339.  
  340. for row in Ntesting:
  341. print(row)
  342. print("Prva mrezha:")
  343. print(correct_first)
  344. print("Vtora mrezha:")
  345. print(correct_second)
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!