import numpy as np
import numpy.random as rnd
import matplotlib.pyplot as plt
import scipy.linalg as lin
import re

data=np.array(map(lambda l: map(float,filter(lambda x: len(x)>0,re.split('\\s+',l))),open('chfdb_chf01_275.txt'))).T
dK=230
pattern=data[1,:dK]
data=data[1,dK:]

def create_mats(dat):
    '''
    create
        A - an initial transition matrix
        pA - pseudocounts for A
        w - emission distribution regression weights
        K - number of hidden states
    '''
    step=5 #adjust this to change the granularity of the pattern
    eps=.1
    dat=dat[::step]
    K=len(dat)+1
    A=np.zeros( (K,K) )
    A[0,1]=1.
    pA=np.zeros( (K,K) )
    pA[0,1]=1.
    for i in xrange(1,K-1):
        A[i,i]=(step-1.+eps)/(step+2*eps)
        A[i,i+1]=(1.+eps)/(step+2*eps)
        pA[i,i]=1.
        pA[i,i+1]=1.
    A[-1,-1]=(step-1.+eps)/(step+2*eps)
    A[-1,1]=(1.+eps)/(step+2*eps)
    pA[-1,-1]=1.
    pA[-1,1]=1.

    w=np.ones( (K,2) , dtype=np.float)
    w[0,1]=dat[0]
    w[1:-1,1]=(dat[:-1]-dat[1:])/step
    w[-1,1]=(dat[0]-dat[-1])/step

    return A,pA,w,K

#initialize stuff
A,pA,w,K=create_mats(pattern)

eta=10. #precision parameter for the autoregressive portion of the model
lam=.1 #precision parameter for the weights prior

N=1 #number of sequences
M=2 #number of dimensions - the second variable is for the bias term
T=len(data) #length of sequences

x=np.ones( (T+1,M) ) # sequence data (just one sequence)
x[0,1]=1
x[1:,0]=data

#emissions
e=np.zeros( (T,K) )
#residuals
v=np.zeros( (T,K) )

#store the forward and backward recurrences
f=np.zeros( (T+1,K) )
fls=np.zeros( (T+1) )
f[0,0]=1
b=np.zeros( (T+1,K) )
bls=np.zeros( (T+1) )
b[-1,1:]=1./(K-1)

#hidden states
z=np.zeros( (T+1),dtype=np.int )

#expected hidden states
ex_k=np.zeros( (T,K) )

# expected pairs of hidden states
ex_kk=np.zeros( (K,K) )
nkk=np.zeros( (K,K) )

def fwd(xn):
    global f,e
    for t in xrange(T):
        f[t+1,:]=np.dot(f[t,:],A)*e[t,:]
        sm=np.sum(f[t+1,:])
        fls[t+1]=fls[t]+np.log(sm)
        f[t+1,:]/=sm
        assert f[t+1,0]==0

def bck(xn):
    global b,e
    for t in xrange(T-1,-1,-1):
        b[t,:]=np.dot(A,b[t+1,:]*e[t,:])
        sm=np.sum(b[t,:])
        bls[t]=bls[t+1]+np.log(sm)
        b[t,:]/=sm

def em_step(xn):
    global A,w,eta
    global f,b,e,v
    global ex_k,ex_kk,nkk

    x=xn[:-1] #current data vectors
    y=xn[1:,:1] #next data vectors predicted from current
    #compute residuals
    v=np.dot(x,w.T) # (N,K) <- (N,1) (N,K)
    v-=y
    e=np.exp(-eta/2*v**2,e)

    fwd(xn)
    bck(xn)

    # compute expected hidden states
    for t in xrange(len(e)):
        ex_k[t,:]=f[t+1,:]*b[t+1,:]
        ex_k[t,:]/=np.sum(ex_k[t,:])

    # compute expected pairs of hidden states    
    for t in xrange(len(f)-1):
        ex_kk=A*f[t,:][:,np.newaxis]*e[t,:]*b[t+1,:]
        ex_kk/=np.sum(ex_kk)
        nkk+=ex_kk

    # max w/ respect to transition probabilities
    A=pA+nkk
    A/=np.sum(A,1)[:,np.newaxis]

    # solve the weighted regression problem for emissions weights
    #  x and y are from above
    for k in xrange(K):
        ex=ex_k[:,k][:,np.newaxis]
        dx=np.dot(x.T,ex*x)
        dy=np.dot(x.T,ex*y)
        dy.shape=(2)
        w[k,:]=lin.solve(dx+lam*np.eye(x.shape[1]), dy)

    #return the probability of the sequence (computed by the forward algorithm)
    return fls[-1]

if __name__=='__main__':
    #run the em algorithm
    for i in xrange(20):
        print em_step(x)

    #get rough boundaries by taking the maximum expected hidden state for each position
    r=np.arange(len(ex_k))[np.argmax(ex_k,1)<3]

    #plot
    plt.plot(range(T),x[1:,0])

    yr=[np.min(x[:,0]),np.max(x[:,0])]
    for i in r:
        plt.plot([i,i],yr,'-r')

    plt.show()

return new NeuralNetConfiguration.Builder()
                .seed(HyperParameter.seed)
                .iterations(HyperParameter.nItr)
                .miniBatch(false)
                .learningRate(HyperParameter.learningRate)
                .biasInit(0)
                .weightInit(WeightInit.XAVIER)
                .momentum(HyperParameter.momentum)
                .optimizationAlgo(
                        OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT  // RMSE: ????
                )
                .regularization(true)
                .updater(Updater.RMSPROP) // NESTEROVS
                // .l2(0.001)
                .list()
                .layer(0,
                        new GravesLSTM.Builder().nIn(HyperParameter.numInputs).nOut(HyperParameter.nHNodes_1).activation("tanh").build())
                .layer(1,
                        new GravesLSTM.Builder().nIn(HyperParameter.nHNodes_1).nOut(HyperParameter.nHNodes_2).dropOut(HyperParameter.dropOut).activation("tanh").build())
                .layer(2,
                        new GravesLSTM.Builder().nIn(HyperParameter.nHNodes_2).nOut(HyperParameter.nHNodes_2).dropOut(HyperParameter.dropOut).activation("tanh").build())
                .layer(3, //"identity" make regression output
                        new RnnOutputLayer.Builder(LossFunctions.LossFunction.MSE).nIn(HyperParameter.nHNodes_2).nOut(HyperParameter.numOutputs).activation("identity").build()) //"identity"
                .backpropType(BackpropType.TruncatedBPTT)
                .tBPTTBackwardLength(100)
                .pretrain(false)
                .backprop(true)
                .build();