"""
...................
Two spirals problem
...................
"""

print(__doc__)

import numpy as np
import matplotlib.pyplot as plt
from sklearn.svm import SVC
from sklearn.neural_network import MLPClassifier
from sklearn.metrics import confusion_matrix
import time

def spirals(n):
    angle=np.linspace(0,n,n)
    angle=np.pi*angle/16
    r=6.5*(104-np.linspace(0,n,n))/104
    x1=r*np.cos(angle)
    x2=r*np.sin(angle)
    A1=np.vstack((x1,x2,np.ones(n,))).T
    A2=np.vstack((-x1,-x2,np.zeros(n,))).T
    data=np.concatenate((A1,A2),axis=0)
    return data

n=100
data=spirals(n)

plt.plot(data[:n,0],data[:n,1],'bo')
plt.plot(data[n:,0],data[n:,1],'go')
plt.show()

X=data[:,:2]
y=data[:,2]

#####################################
# Feladat: probaljuk meg elvalasztani a ket halmazt MLP-vel illetve SVM-mel (RBF kernel)!
# Az eredmenyt szemleltessuk az RBF_iris kod vegen latott modon, 
         # azaz egy racs felvetelevel, ahol minden racspontot beszinezunk a halozat altal nyujtott ertek alapjan!
# Legyen kulon abra MLP-re es SVM-re
# mlp-nel hasznalhatjuk a halozat .predict_proba attributumat


##############################################

svm=SVC(gamma=1,C=1)                 # C: jegyzetben lambda, regularizációs paraméter
start_time = time.time()
svm.fit(X,y)
print('Ennyi ideig tartott SVM-mel: {0} mp'.format(time.time()-start_time))

##############################################

mlp = MLPClassifier(hidden_layer_sizes=(10,10,10,10), max_iter=100,solver='lbfgs',learning_rate_init=.1,activation='tanh')
start_time = time.time()
mlp.fit(X,y)
print('Ennyi ideig tartott MLP-vel: {0} mp'.format(time.time()-start_time))

##############################################

x1_test, x2_test = np.meshgrid(np.linspace(-7,7,100), np.linspace(-7,7,100)) 
Z = svm.decision_function(np.c_[x1_test.ravel(), x2_test.ravel()])  # decision_function: távolság az elválasztó hipersíktól
Z = Z.reshape(x1_test.shape)

plt.pcolormesh(x1_test, x2_test, Z)
plt.scatter(X[:,0], X[:,1],c=y,edgecolors='1')
plt.show()

x1_test, x2_test = np.meshgrid(np.linspace(-7,7,100), np.linspace(-7,7,100)) 
test=np.vstack((x1_test.ravel(),x2_test.ravel())).T
Z=mlp.predict_proba(test[:,])[:,1]                        # predict_proba: valsegek osztalyonkent

Z = Z.reshape(x1_test.shape)

plt.pcolormesh(x1_test, x2_test, Z)
plt.scatter(X[:,0], X[:,1], c=y,edgecolors='1')
plt.show()

######################################

# Tévesztési mátrix:

y_true = y
y_pred = mlp.predict(X)
print(confusion_matrix(y_true, y_pred))
print('Pontosság a tanítóhalmazon MLP-vel: {0}%'.format(mlp.score(X, y)*100))