import numpy as np
import matplotlib.pyplot as plt

alpha=0.01
niters=1000

def h(X, theta):
    return 1 / (1 + np.e ** -(X.dot(theta.T)))


def J(X, y, theta):
    m = X.shape[0]
    y_hat = h(X, theta)
    erro = (-y * np.log(y_hat) - (1-y) * np.log(1-y_hat)).sum(0)
    return erro / m


def GD(X, y, theta, alpha, niters):
    m = X.shape[0]
    cost = np.zeros((niters,1))
    print('iteração:')
    for k in range(0, niters):
        print(' ',k,end='')
        y_hat = h(X, theta)
        erro = ((y_hat-y) *X).sum(0)/m
        theta -= (alpha * (erro))
        cost[k] = J(X, y, theta)
        print('\r\r\r\r\r\r',end='')
    return (cost, theta)

def featureScaling(X):
  X=X-np.min(X,0)
  den=np.max(X,0)-np.min(X,0)
  return X/den

def getClass(y):
    lem=((np.max(y)-np.min(y))/2 + np.min(y))
    y = y >= lem
    return y

def getAll(y1,y2):
    trueNegative = 0
    falseNegative = 0
    truePositive = 0
    falsePositive = 0
    k = 0
    for i in y1:
        if i == 0 and i == y2[k]:
            trueNegative+= 1
        elif i == 1 and i == y2[k]:
            truePositive+= 1
        elif i == 0 and i != y2[k]:
            falseNegative+= 1
        elif i == 1 and i != y2[k]:
            falsePositive+= 1
        k+=1
    return trueNegative, truePositive, falseNegative, falsePositive


f=open('Skin_NonSkin.txt')
## delete \n at the end of the string
data=[d[:-1] for d in f.readlines()]
## split the string into the values
data=[d.split('\t') for d in data]
np.random.shuffle(data)
# now data is a list of lists
X=[d[:-1] for d in data]

X=np.array(X,dtype=int)









#### Inicio da área que é permita alguma mudança

X=featureScaling(X)
X=np.insert(X,0,1,axis=1)

#### Fim da área que é permitida alguma mudança





y=[d[-1] for d in data]
y=np.reshape(np.array(y,dtype=int),(len(y),1))
y[y==5]=0
y[y==6]=1

Theta=np.zeros((1,X.shape[1]))

nsize=int(X.shape[0]*.7)

Xtr=X[:nsize,:]
Xte=X[nsize:,:]
ytr=y[:nsize]
yte=y[nsize:]

c,t=GD(Xtr,ytr,Theta,alpha,niters)

y_hat=getClass(h(Xtr,t))
print('Taxa de acerto (treino):', np.mean(y_hat==ytr))
y_hat=getClass(h(Xte,t))
print('Taxa de acerto (teste):', np.mean(y_hat==yte))



print('Blaaa\n', y_hat)

y_hat[y_hat==False] = 0
y_hat[y_hat==True] = 1

print('Blaaa22222222\n', y_hat)

print('bleeeee\n', yte)

yte[yte==2] = 0
yte[yte==1] = 1

print('bleeeee2222\n', yte)

trueNegative, truePositive, falseNegative, falsePositive=getAll(y_hat,yte)


print(trueNegative, truePositive, falseNegative, falsePositive)

print('precisao -> ' + str(truePositive / (truePositive + falsePositive)))
print('recall -> ' + str(truePositive / (truePositive + falseNegative)))
print('acuracia -> ' + str((truePositive + trueNegative) / (truePositive + falsePositive + trueNegative + falseNegative)))

##plt.plot(c)
##plt.show()