test_conv_lstm

import numpy as np

import IPython

# transition model and observation model
A = np.array([[0.2, 0.8, 0], [0., 0.2, 0.8], [0, 0., 1]])
O = np.array([[0.8, 0.2, 0], [0.1, 0.8, 0.1], [0.0, 0.2, 0.8]])
# O = O.transpose()

# initial setup
pi = np.array([1, 0, 0])
pi = np.transpose(pi)

# sequence
S = [1, 0, 1, 2]

# target: argmax P{{Zi} | pi, A, O, S}
#            Xi

bt = []

for step, s in enumerate(S[1:]):
    # for every state, find the most possible ancestor by max func
    temp = []
    bt_row = []
    for i in range(len(pi)):
        a = A[:, i] # the array of possible states to i state probability
        temp.append(np.max(pi*a))
        bt_row.append(np.argmax(pi*a))
        # print bt_row

    pi = temp
    print 'step {}: the predicted state based on transition:\n {}'.format(step, pi)

    # P{X_k|O_k} = P{O_k = s|X_k}P{X_k}   and P{X_k} = pi
    print 'observation:'+str(s)
    ita = 0
    for i in range(len(pi)):
        pi[i] = O[i, s]*pi[i]
        # print O[i, s]
        ita += pi[i]

    pi = pi/ita
    print 'step {}: the predicted state updated with observation:\n {}'.format(step, pi)
    # print bt_row

    bt.append(bt_row)
    # IPython.embed()
    print '===================================================='

# IPython.embed()

print 'backtracing matrix:\n {}'.format(bt)

# backtracing
print '----------------------------------------------'
print 'start to backtrace from the max at terminal...'
Ind_T = np.argmax(pi)
ind = Ind_T
print 'the state at termination is: {}'.format(ind)


# IPython.embed()

for j in np.flip(range(len(bt))):
    print 'step {}:'.format(j)
    print 'the state backwards are: {}'.format(bt[j][ind])
    ind = bt[j][ind]