def split_into_folds(self, inpL, classif,save=True, excludeL=None,info=''):
out_allinfo={}
out_lengths={}
numSegm=Constants.numSegm_total # one of the segments will be validation and the rest used in training
for seg in range(numSegm):
out_allinfo[seg]={}
out_lengths[seg]=0
for scafName,text,fetureV,seq in inpL:
randomSegment=np.random.randint(numSegm)
out_allinfo[randomSegment][scafName]={'text':text,'seq':seq,'len':len(seq),'features':fetureV} #
out_lengths[randomSegment]+=len(seq)
if self.verb>0:
print(' achieved split for %d scaffolds to segments:'%len(inpL),[ (x,len(out_allinfo[x]),out_lengths[x]) for x in out_allinfo ])
for seg in range(numSegm):
sum=out_lengths[seg]/1.e6
print('seg:',seg,' size=%.2f (MB) numScaf=%d'%(sum,len(out_allinfo[seg])))
out_allinfo['info']=info
# save segments
if save==True:
out_file=self.trainvalid_1hot_dataPath+'/'+self.name+'.%s-scaff-split.yml'%classif
write_yaml(out_allinfo,out_file)
return out_allinfo
def split_species(self, inpL, role, save=True, excludeL=None, info=''):
out0 = {}
out1 = {}
numSegm = self.numSegm + 1 # the last one is not used in any training
for seg in range(numSegm):
out0[seg] = {}
out1[seg] = 0
for scafN, text, fetureV, seq in inpL:
ix = np.random.randint(numSegm)
out0[ix][scafN] = {
'text': text,
'seq': seq,
'len': len(seq),
'features': fetureV
} #
out1[ix] += len(seq)
if self.verb > 0:
print(
' achieved split for %d scaffolds to segments:' % len(inpL),
[(x, len(out0[x]), out1[x]) for x in out0])
for seg in range(numSegm):
sum = out1[seg] / 1.e6
print('seg:', seg,
' size=%.2f (MB) numScaf=%d' % (sum, len(out0[seg])))
out0['info'] = info
# save segments
if save == True:
outF = self.dataPath + '/' + self.name + '.%s-scaff-split.yml' % role
write_yaml(out0, outF)
return out0
def save_training_history(self):
outD = self.train_hirD
outF = self.outPath + '/' + self.name + '.history.yml'
write_yaml(outD, outF)
print(
cnt['inp'], scaffN, 'decision=%s, avr score:' % decision,
classif_avrg_score_str,
' len=%.1fk samples=%d' % (len(sequenceString) / 1000., len(sampList)))
if args.verb > 0:
print(Yscores_samples, classif_avrg_score_str, classif_score,
scores_samples_list_class1, scores_samples_list_class0,
scores_scaffs_list, Yclass, sampListHotEncodedA,
featureListXsamplesA)
classif_details_yaml['score'] = classif_score
classif_details_yaml['model_info'] = deep.info
#print('out classif_details_yaml='); pprint(classif_details_yaml)
#write under outPR the info for this scaffold prediction in a yaml file
write_yaml(classif_details_yaml,
args.outPath + '/%s.assayer.yml' % (scaffN), 0)
nClass = cnt['plasmid'] + cnt['main'] + cnt['ambig']
#print('M:%s endCnt:'%given_class,cnt,' fraction: Plasm=%.3f Ambig=%.3f Main=%.3f'%(cnt['plasmid']/nClass,cnt['ambig']/nClass,cnt['main']/nClass))
print('Counts: Plasm=%s Ambig=%s Main=%s nCount=%s' %
(cnt['plasmid'], cnt['ambig'], cnt['main'], nClass))
f_predix.close()
# make plot of all scores
#ROC curve
#print("Yclass %s " % (Yclass))
#print("scores_scaffs_list %s " % (scores_scaffs_list))
####Print the AUC-ROC and FP-TP rate list to a file
if 1 in Yclass and 0 in Yclass: ###Need both classes for this to be meaningful
fpr, tpr, _ = roc_curve(