def ecrireXml(visu, p):
for plate in visu:
for well in visu[plate]:
if p==None:
fichier = open("PL"+plate+"___P"+well+"___T00000.xml", 'w')
else:
fichier = open(str(p)+"PL"+plate+"___P"+well+"___T00000.xml", 'w')
fichier.write(fHacktrack2.initXml())
compteur=1
for index in visu[plate][well]:
for c in visu[plate][well][index]:
result = "\n <Marker_Type>\n <Type>{0}</Type>".format(compteur)
numFramePourXml = index+1
result+="\n <Marker>\n <MarkerX>{1}</MarkerX>\n <MarkerY>{2}</MarkerY>\n <MarkerZ>{0}</MarkerZ>\n </Marker>".format(numFramePourXml, c[0], c[1])
result+="\n </Marker_Type>"
fichier.write(result)
fichier.write(fHacktrack2.finirXml())
return
def trajPreparation(dicT, training):
r={}; coord={}; nuCount=[0 for x in range(len(moments))]
for plate in dicT:
if plate not in r:
r[plate]={}
coord[plate]={}
for well in dicT[plate]:
if well not in r[plate]:
r[plate][well]=[]
coord[plate][well]=[]
print plate, well
dicC = dicT[plate][well]
fichier=open("PL{0}___P{1}___T00000.xml{2}".format(plate, well, training), 'w'); fichier.write(fHacktrack2.initXml())
tracklets=[]
compteur = 1
#signature de featuredTraj.__init__ : (self, ide, labelsSequence, beginning, end, num, features, to=None, fr=None):
#going into all trajectories to get small tracklets
#AND keeping in memory which tracklet they're related to
#once we have all tracklets we can do the findCorresp
#FINALLY we can recoller les morceaux and get model accuracy/trajectory instead of /movement from t to t+1
print 'Pass of all tracks'
for traj in dicC.lstTraj:
t=traj.lstPoints.keys(); t.sort();
# if traj.numCellule ==36:
# fichier.write(fHacktrack2.ecrireXml(36, traj))
print "une trajectoire", traj.numCellule
print "merge :", traj.fusion
print "split :", traj.mitose
print 'debut de la trajectoire :', min(t), "fin ", max(t)
#CHECK qu'est-ce que min(t), max(t)
#PLUS compute to and from for all tracklets
#Here we don't want to delete all that happens after a merge
#we want to cut the trajectory into pieces between all split and merge
events = list(traj.fusion); events.extend(traj.mitose); events.sort(reverse=True)
events = filter(lambda x: x>min(t)[0] and x<max(t)[0], events)[0]
# if traj.fusion !=-1:
# if min(traj.fusion)<min(t):
# continue
# traj=traj.copyBefore(min(traj.fusion))
# t=sorted(traj.lstPoints.keys())
#
# print 'apres le check merge, plus de merge normalement : ', traj.fusion , 'et que des splits avant : ', traj.mitose
print "evenements de la trajectoire ", events
if events !=[]:
end = max(t)[0]
for event in events:
# if end-mit>9:
print "event ", event
track = traj.copyBetween(event, end)
r[plate][well].append(trajFeatures.featuredTraj(compteur, labelsSeq(track), track.numCellule, None, to, fr))
compteur +=1
#tracklets.append()
end = event
track= traj.copyBefore(end)
r[plate][well].append(trajFeatures.featuredTraj(compteur, labelsSeq(track), track.numCellule, None, to, fr))
compteur +=1
#tracklets.append()
else:#if end-t[0][0]>9:
track = traj
r[plate][well].append(trajFeatures.featuredTraj(compteur, labelsSeq(track), track.numCellule, None, to, fr))
compteur +=1
#tracklets.append(traj)
#ATTENTION here we throw away tracklets whose length is smaller than 10 (otherwise we don't have diffusion info because
#I use delta t in range (1, len(track)/3 +1) ie (1,2,3) if len(track)>9, and then do regression on it,
#so if we have less than three points to do linear regression it seems a bit dubious to me)
# if not len(tracklets)==len(filter(lambda x: len(x.lstPoints)>9 ,tracklets)):
# raise
print 'Second pass of all tracks to extract features of tracks longer than 9 frames'
#euh no here we're not doing this
# pbl=dict(zip(moments, [[] for x in range(len(moments))]))
# for track in filter(lambda x: len(x.lstPoints)>9 ,tracklets):
# t=track.lstPoints.keys(); t.sort();
# labelsSequence = [k[1] for k in t]
# x= [track.lstPoints[k][0] for k in t]
# y =[track.lstPoints[k][1] for k in t]
# #pdb.set_trace()
# coord[plate][well].append([x, y])
# print "debut de l'extrait de trajectoire", min(t)[0], "fin ", max(t)[0]
#
# #Pour l'instant on ne normalise pas par rapport au deplacement moyen sur toute l'image car on ne considere qu'une partie de l'image
# f=computingFeatures(track, movie_length[plate][well], np.mean(average, 0))
# if 'pbl' in f:
# nu=f['pbl']
# #nuCount[nu].append(len(track.lstPoints))
# pbl[nu].append(track)
# fichier.write(fHacktrack2.ecrireXml(nu, track))
# r[plate][well].append(featuredTraj(compteur, labelsSequence, min(t)[0], max(t)[0], track.numCellule, f))
# if track.numCellule ==12:
# fichier.write(fHacktrack2.ecrireXml(12, traj))
# compteur +=1
##
# fichier.write(fHacktrack2.finirXml()); fichier.close()
# f=open('pblCoeffCorr{0}.pkl'.format(training), 'w')
# pickle.dump(pbl, f); f.close()
return r, coord
