
We load the sequence and the
lexique corresponding to the HMM.
import sequence
s=sequence.Sequence(fic="lambda.seq")
len(s)
import lexique
lx=lexique.Lexique(fprop="lprop1")
print lx
 We build and draw the
partition of the states of the
most probable "path" computed with Viterbi algorithm.
import partition
p=partition.Partition()
p.viterbi(s,lx)
len(p)
print p
p.draw_nf("lambda1.ps",num=1)
 We compute the Matrice of the
logprobabilities of the descriptors given the sequence and the
model, by ForwardBackward algorithm.
import matrice
m=matrice.Matrice()
m.fb(s,lx)
print m[:10]
 We build a new partition, in which each segment is made where the most
probable descriptor in former Matrice is the same on a continuous set
of positions.
p2=partition.Partition()
p2.read_Matrice(m)
len(p2)
 And if we want to draw the partition such that the height of
each arc is proportional with minus the density of logprobability
of the model on this segment.
p2.draw_nf("lambda_val.ps",num=1,func=lambda x: x.val()/len(x))
 And we want to draw only the segments described by descriptor 3.
p2.draw_nf("lambda_val3.ps",seg=[3],func=lambda x: x.val()/len(x))
 We compute the proportion of common descriptions between both
partitions.
float(p2.pts_comm(p))/len(s)
 We build the 50partitioning from the predictions of the
lexique, and draw it.
import parti_simp
ps=parti_simp.Parti_simp()
ps.mpp(s,lx,50)
ps.draw_nf("lambda_ps.ps",num=1)
 We compute the list of the similarities between the partitions
of ps and both partitions p1 and p2.
l=[]
for i in range(len(ps)):
l.append([i+1,float(ps[i].pts_comm(p))/len(s),\
float(ps[i].pts_comm(p2))/len(s)])
for i in l:
print i
Notice and compare the best scores and their corresponding number of
segments with the actual number of segments of p and
p2.