SHARE
TWEET

Utility

a guest Jun 15th, 2019 94 Never
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
  1. import math
  2. from sklearn.preprocessing import MinMaxScaler
  3. from sklearn.cluster import KMeans
  4. from sklearn.neighbors import NearestNeighbors
  5. import numpy as np
  6. from tqdm import tqdm
  7. import matplotlib.pyplot as plt
  8. import Layer
  9. import Network
  10.  
  11.  
  12. # numer funkcji,pierwszy argument,drugi argument - opcjonalny
  13. def getFunctionValue(choice, x1, x2=0):
  14.     if (choice) == 1:
  15.         return math.sqrt(x1)
  16.     elif (choice == 2):
  17.         return math.sin(x1)
  18.     else:
  19.         return math.sin(x1 * x2) + math.cos(3 * (x1 - x2))
  20.  
  21.  
  22. # liczba punktow,zakres,liczba argumentow
  23. def getInputs(pointsNumber, pointsRange, argumentsNumber):
  24.     inputs = np.asarray(np.full((pointsNumber, argumentsNumber), pointsRange[0], dtype=float))
  25.     diff = pointsRange[1] - pointsRange[0]
  26.     j = diff / (pointsNumber - 1)
  27.     for counter, row in enumerate(inputs):
  28.         row += j * counter
  29.     if (argumentsNumber == 2):
  30.         inputs[:, 1] = np.flip(inputs[:, 1])
  31.     return np.asmatrix(inputs)
  32.  
  33.  
  34. # tablica argumentow,numer funkcji
  35. def getOutputs(inputs, functionNumber):
  36.     rows, columns = inputs.shape
  37.     outputs = np.asmatrix(np.zeros((rows, 1), dtype=float))
  38.  
  39.     if (columns == 1):
  40.         for i in range(rows):
  41.             outputs[i, 0] = getFunctionValue(functionNumber, inputs[i, 0])
  42.     else:
  43.         for i in range(rows):
  44.             outputs[i, 0] = getFunctionValue(functionNumber, inputs[i, 0], inputs[i, 1])
  45.     return outputs
  46.  
  47.  
  48. # DZIELENIE NA WEJSCIE I WYJSCIA ZBIORU TESTOWEGO I UCZACEGO
  49. def divideDataset(inputs, outputs):
  50.     height = inputs.shape[0]
  51.     x = int(round(0.3 * height))
  52.     y = int((inputs.shape[0] / x) + 1)
  53.  
  54.     inputTrain = np.delete(inputs, np.arange(0, inputs.size, y), 0)
  55.     outputTrain = np.delete(outputs, np.arange(0, outputs.size, y), 0)
  56.  
  57.     inputTest = inputs[::y]
  58.     outputTest = outputs[::y]
  59.  
  60.     return inputTrain, outputTrain, inputTest, outputTest
  61.  
  62.  
  63. def normalize(dataset):
  64.     rows, columns = dataset.shape
  65.     normalized = np.asmatrix(np.zeros((rows, columns), dtype=float))
  66.     for i in range(columns):
  67.         maxCol = max(dataset[:, i])
  68.         minCol = min(dataset[:, i])
  69.         for j in range(rows):
  70.             normalized[j, i] = (dataset[j, i] - minCol) / (maxCol - minCol)
  71.     return normalized
  72.  
  73.  
  74. def learn(network, inputMatrix, outputMatrix, epochs):
  75.     rows = inputMatrix.shape[0]
  76.     arr = list(range(0, rows))
  77.     plt.ion()
  78.     plt.figure()
  79.     x = [None] * rows
  80.     results = [None] * rows
  81.     expected = [None] * rows
  82.  
  83.     for i in tqdm(range(epochs)):
  84.         for j in tqdm(range(rows)):
  85.             network.simulate(inputMatrix[arr[j]].T, outputMatrix[arr[j]].T)
  86.             x[arr[j]] = inputMatrix[arr[j], 0]
  87.             expected[arr[j]] = outputMatrix[arr[j], 0]
  88.             results[arr[j]] = network.layers[-1].output[0, 0]
  89.  
  90.         plot2D(x, results, expected, i, 'learn')
  91.         np.random.shuffle(arr)
  92.  
  93.  
  94. def test(network, input, output):
  95.     rows = input.shape[0]
  96.     x = list()
  97.     results = list()
  98.     expected = list()
  99.  
  100.     for i in range(rows):
  101.         network.forwardPropagate(input[i].T)
  102.         x.append(input[i, 0])
  103.         results.append(network.layers[-1].output[0, 0])
  104.         expected.append(output[i, 0])
  105.  
  106.     plot2D(x, results, expected, 1, 'test')
  107.     plt.pause(25)
  108.  
  109.  
  110. def plot2D(x, results, expected, epochNumber, mode):
  111.     plt.clf()
  112.     plt.xlabel('X')
  113.     plt.ylabel('Y')
  114.     plt.plot(x, results, 'g', label='Approximation')
  115.     plt.plot(x, expected, 'r', label='Function')
  116.     plt.legend()
  117.     if (mode == 'learn'):
  118.         plt.title('{}\nepoch {}'.format('current', epochNumber))
  119.     plt.pause(0.0001)
  120.  
  121.  
  122. #
  123. # def plot3D(x1, x2, results, expected, epochNumber, mode):
  124.  
  125.  
  126. def getCentroids(centroidsNumber, data):
  127.     kmeans = KMeans(n_clusters=centroidsNumber)
  128.     kmeans.fit(data)
  129.     return np.asmatrix(kmeans.cluster_centers_)
  130.  
  131.  
  132. def getWidths(centroids):
  133.     widths = np.asmatrix(np.zeros((centroids.shape[0], 1), dtype=float))
  134.     knn = NearestNeighbors(n_neighbors=5, algorithm='auto')
  135.     knn.fit(centroids)
  136.     # zwraca odleglosc od sasiadow i ich indeksy
  137.     distances, indexes = knn.kneighbors(centroids)
  138.     for counter, row in enumerate(distances):
  139.         sum = 0
  140.         for i in range(1, knn.n_neighbors):
  141.             sum += pow(row[i], 2)
  142.         r = math.sqrt(sum / (knn.n_neighbors - 1))
  143.  
  144.         widths[counter] = r;
  145.     return widths
  146.  
  147.  
  148. def createNetwork(inputMatrix, outputMatrix, hiddenLayers, layersType):
  149.     layers = np.array([])
  150.  
  151.     if (layersType[0] == 'rbf'):
  152.         centroids = getCentroids(hiddenLayers[1], inputMatrix)
  153.         widths = getWidths(centroids)
  154.  
  155.         temp = np.array([Layer.rbfLayer(centroids, widths)])
  156.         layers = np.append(layers, temp)
  157.  
  158.     elif (layersType[0] == 'affine'):
  159.         temp = np.array([Layer.affineLayer(hiddenLayers[1], inputMatrix.shape[1], 1, 'sig')])
  160.         layers = np.append(layers, temp)
  161.  
  162.     for i in range(1, hiddenLayers[0]):
  163.         if (layersType[i] == 'rbf'):
  164.             centroids = getCentroids(hiddenLayers[i + 1], inputMatrix)
  165.             widths = getWidths(centroids)
  166.  
  167.             temp = np.array([Layer.rbfLayer(centroids, widths)])
  168.             layers = np.append(layers, temp)
  169.  
  170.         elif (layersType[i] == 'affine'):
  171.             temp = np.array([Layer.affineLayer(hiddenLayers[i + 1], hiddenLayers[i], 1, 'sig')])
  172.             layers = np.append(layers, temp)
  173.  
  174.     temp = np.array([Layer.affineLayer(outputMatrix.shape[1], hiddenLayers[-1], 1, 'lin')])
  175.     layers = np.append(layers, temp)
  176.  
  177.     network = Network.Network(layers)
  178.     return network
RAW Paste Data
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!
 
Top