r = [a, b, c] r = np.min(r) # Only the values from the DF are useful in this analysis. When plotting a # matrix graph, the DF come in handy. cl1 = cl1.values[:r, :] cl2 = cl2.values[:r, :] cl3 = cl3.values[:r, :] # Our input x = np.r_[cl1, cl2, cl3] x = np.array(x, dtype=float) x = conv.normalize(x) # Our output using a proprietary function (See pyconv.py). y = conv.out_b([a, b, c]) # Test parameters. Desired train ration an Nearest Neighbors are altered here. trainArr = [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9] nnArr = [1, 5, 7, 11, 13, 17] testN = 30 # Overwrite Results? ovrw = 1 # Initializing our algorithm (See algorithm.py). knn = alg.KNearestNeighbors(x, y) caMax, caMin, caMean, caStd = list(), list(), list(), list() # For loop to calculate KNN based on trainArr and nnArr for nnNumber in nnArr:
x = np.r_[cl1, cl2, cl3]
x = np.array(x, dtype=float)
#y = conv.out_b([950,950,950])
x = (x - np.min(x, 0)) / (np.max(x, 0) - np.min(x, 0))
caMax, caMin, caMean, caStd = list(), list(), list(), list()
xx = list()
t_size = 5
for n in range(0, int(len(x) / t_size)):
xx.append(np.mean(x[n * t_size:(n + 1) * t_size, :], 0))
m = int(len(x) / (3 * t_size))
xx = np.array(xx, dtype=float)
y = conv.out_b([m, m, m])
knn = alg.KNearestNeighbors(xx, y)
trainArr = [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]
nnArr = [1, 5, 7, 11, 13, 17]
for nnNumber in nnArr:
for trainSize in trainArr:
confArrayList = list()
for n in range(0, 40):
knn.train(nnNumber, trainSize)
confArrayList.append(knn.confArray)
cal = np.array(confArrayList)
caMax.append(np.diag(np.max(cal, 0)))
caMin.append(np.diag(np.min(cal, 0)))
import pyconv as conv
def s_anal(cal):
caMax = np.max(cal, 0)
caMin = np.min(cal, 0)
caMean = np.mean(cal, 0)
caStd = np.std(cal, 0)
return caMax, caMin, caMean, caStd
irisdf = pd.read_csv('data/iris.csv')
iris = np.array(irisdf.values)
x = np.array(iris[:, 0:4], dtype='float64')
y = conv.out_b([50, 50, 50])
x = conv.normalize(x)
inputLayer = 4
hiddenLayer = 4
outputLayer = 3
learningRate = 0.3
regularizationParameter = 0
net = alg.MultiLayerPerceptron(inputLayer, hiddenLayer, outputLayer,
learningRate, regularizationParameter)
net.initialize(x, y)
caMax, caMin, caMean, caStd = list(), list(), list(), list()
a, b, c = cl1.shape[0], cl2.shape[0], cl3.shape[0] r = [a, b, c] r = np.min(r) # Only the values from the DF are useful in this analysis. When plotting a # matrix graph, the DF come in handy. cl1 = cl1.values[:960, :] cl2 = cl2.values[:960, :] cl3 = cl3.values[:960, :] # Our input x = np.r_[cl1, cl2, cl3] x = np.array(x, dtype=float) x = conv.normalize(x) # Our output using a proprietary function (See pyconv.py). y = conv.out_b([a, b, c]) # Overwrite Results? ovrw = 1 # Test parameters. Desired train ration an Nearest Neighbors are altered here. lenghtArr = np.linspace(1, 80, 80, dtype=int) nnArr = [1] testN = 30 trainSize = 0.5 numC = 3 # Initializing our algorithm (See algorithm.py). (m, n) = x.shape knn = alg.KNearestNeighbors(x, y)
#a,b,c = 50,50,50
#cl1 = iris[0:a,:]
#cl2 = iris[a:a+b,:]
#cl3 = iris[a+b:a+b+c,:]
cl3 = np.array(cl3.values[:, 0:4])
cl2 = np.array(cl2.values[:, 0:4])
cl1 = np.array(cl1.values[:, 0:4])
cl3 = cl3[0:960, :]
cl2 = cl2[0:960, :]
cl1 = cl1[0:960, :]
x = np.r_[cl1, cl2, cl3]
y = conv.out_b([a, b, c])
x = (x - np.min(x, 0)) / (np.max(x, 0) - np.min(x, 0))
t_size = 10
xx = list()
for n in range(0, int(len(x) / t_size)):
xx.append(np.mean(x[n * t_size:(n + 1) * t_size, 0:5], 0))
m = int(len(x) / (3 * t_size))
xx = np.array(xx, dtype=float)
y = conv.out_b([m, m, m])
dist = 'eDist'
km = alg.kmeans(xx, y)