def saveInfo(c):
tDir = d['-o'][1] + "/" + "Arch_" + str(c + 1)
try:
# Save HTML file for best model of each fold an architecture
os.mkdir(tDir)
for fold in range(d['-kfold']):
td = tDir + "/Fold_" + str(fold + 1)
os.mkdir(td)
sf.saveDetails(toArr[c][fold], td, d['-i'], [], d['-tss'], 1, d['-plotExtra'], d['-pCol'], d['-sortBy'], d['-eps'])
except OSError:
print "ERROR: Cannot create directory in", d['-o'][1]
exit(2)
if c == 0 or d['-v'] == 0: return
for fold in range(d['-kfold']):
# Save likelihood plot if -v flag is set
for sd in range(d['-lcount']):
if os.path.isfile(d['-o'][1] + "/." + str(c + 1) + "_" + str(fold + 1) + "_" + str(sd)):
os.system("mv" + " " + d['-o'][1] + "/." + str(c + 1) + "_" + str(fold + 1) + "_" + str(sd) + " " + tDir + "/Fold_" + str(fold + 1) + "/" + plotFileHiddenName)
os.system("gnuplot" + " " + "-e" + " " + "'filename=\"" + tDir + "/Fold_" + str(fold + 1) + "/" + plotFileHiddenName + "\"; var=\"" + tDir + "/Fold_" + str(fold + 1) + "/" + plotLikelihoodImage + "\"'" + " " + d['-v'][0])
if len(d['-v']) != 1: os.system("gnuplot" + " " + "-e" + " " + "'filename=\"" + tDir + "/Fold_" + str(fold + 1) + "/" + plotFileHiddenName + "\"; var=\"" + tDir + "/Fold_" + str(fold + 1) + "/" + plotLikelihoodImageEPS + "\"'" + " " + d['-v'][1])
os.system("rm" + " " + "-f" + " " + tDir + "/Fold_" + str(fold + 1) + "/" + plotFileHiddenName)
def saveInfo(c):
tDir = d['-o'][1] + "/" + "Arch_" + str(c + 1)
try:
# Save HTML file for best model of each fold an architecture
os.mkdir(tDir)
for fold in range(d['-kfold']):
td = tDir + "/Fold_" + str(fold + 1)
os.mkdir(td)
sf.saveDetails(toArr[c][fold], td, d['-i'], [], d['-tss'], 1, d['-plotExtra'], d['-pCol'], d['-sortBy'], d['-eps'])
except OSError:
print "ERROR: Cannot create directory in", d['-o'][1]
exit(2)
if c == 0 or d['-v'] == 0: return
for fold in range(d['-kfold']):
# Save likelihood plot if -v flag is set
for sd in range(d['-lcount']):
if os.path.isfile(d['-o'][1] + "/." + str(c + 1) + "_" + str(fold + 1) + "_" + str(sd)):
os.system("mv" + " " + d['-o'][1] + "/." + str(c + 1) + "_" + str(fold + 1) + "_" + str(sd) + " " + tDir + "/Fold_" + str(fold + 1) + "/" + plotFileHiddenName)
os.system("gnuplot" + " " + "-e" + " " + "'filename=\"" + tDir + "/Fold_" + str(fold + 1) + "/" + plotFileHiddenName + "\"; var=\"" + tDir + "/Fold_" + str(fold + 1) + "/" + plotLikelihoodImage + "\"'" + " " + d['-v'][0])
if len(d['-v']) != 1: os.system("gnuplot" + " " + "-e" + " " + "'filename=\"" + tDir + "/Fold_" + str(fold + 1) + "/" + plotFileHiddenName + "\"; var=\"" + tDir + "/Fold_" + str(fold + 1) + "/" + plotLikelihoodImageEPS + "\"'" + " " + d['-v'][1])
os.system("rm" + " " + "-f" + " " + tDir + "/Fold_" + str(fold + 1) + "/" + plotFileHiddenName)
def getLabels(d): # Assign labels based on input model
dirname = d['-o'][1]
features = getFeatures(d['-f'])
printLearnDetails(d['-o'])
m = ld.learn(d['-f'], dirname + "/" + tempLabelsFile, d['-m'], features)
print "\nModel learnt successfully.\nSaving details..."
sf.saveDetails(m, dirname, d['-i'], [], d['-tss'], 0, d['-plotExtra'], d['-pCol'], d['-sortBy'], d['-eps'])
del features, m
print "Goodbye!"
gc.collect()
return dirname
def getLabels(d): # Assign labels based on input model
dirname = d['-o'][1]
features = getFeatures(d['-f'])
printLearnDetails(d['-o'])
m = ld.learn(d['-f'], dirname + "/" + tempLabelsFile, d['-m'], features, dirname + "/" + learnScoresFile)
print "\nModel learnt successfully.\nSaving details..."
sf.saveDetails(m, dirname, d['-i'], [], d['-tss'], 0, d['-plotExtra'], d['-pCol'], d['-sortBy'], d['-eps'])
os.system("cp " + dirname + "/" + clusterDetailsFile + " " + dirname + "/" + clusterDetailsFile + "1")
os.system("paste " + dirname + "/" + clusterDetailsFile + "1 " + dirname + "/" + learnScoresFile + " > " + dirname + "/" + clusterDetailsFile)
os.system("rm " + dirname + "/" + learnScoresFile)
os.system("rm " + dirname + "/" + clusterDetailsFile + "1")
del features, m
print "Goodbye!"
gc.collect()
return dirname
def getModel(d): # Learn best model
dirname = d['-o'][1]
n = mp.cpu_count()
if d['-proc'] < n and d['-proc'] > 0: n = d['-proc']
if 3*d['-kfold']*d['-lcount'] < n: n = 3*d['-kfold']*d['-lcount']
d['-proc'] = n
features = getFeatures(d['-f'])
if d['-tss'] == 0: d['-tss'] = features/2
if d['-tss'] > features:
print "ERROR: -tss is more than the length of sequences"
exit(1)
count = d['-maxarch'] - d['-minarch'] + 1
printDetails(d['-o'], count)
saveSettings(d)
m, cvals = ev.learn(d, dirname + "/" + tempLabelsFile, count)
print "\n\nModel learnt successfully.\nSaving details..."
sf.saveDetails(m, dirname, d['-i'], cvals, d['-tss'], 0, d['-plotExtra'], d['-pCol'], d['-sortBy'], d['-eps'])
print "Goodbye!"
gc.collect()
return dirname
def learn(dt, outfile, count):
global d, ds, trainSets, testSets, lcount
lcount = dt['-lcount']
ds = getData(libctest.getData(dt['-f'], outfile))
pos = libctest.posList(ds.contents.n)
# Learn best model directly if -minarch and -maxarch are same
# if dt['-maxarch'] == dt['-minarch']:
# d = dt
# printBestModel(dt['-maxarch'])
# m = learnModel(dt['-maxarch'], dt['-o'][1] + "/" + tempFile)
# return m, []
for i in range(dt['-kfold']): # Get randomized train sets and test sets for every fold
trainSets.append(getData(libctest.getTrainSubset(ds, i, dt['-kfold'], pos)))
testSets.append(getData(libctest.getTestSubset(ds, i, dt['-kfold'], pos)))
if dt['-lambda'] != -1: # Learn models by single lambda value
m, cvals = learnDiffLambda(dt, outfile, count, ds, trainSets, testSets)
else: # Learn models by varying lambda
print "\n\nTrying Lambda", 0, "\n\n"
finalOut = dt['-o'][1]
dt['-lambda'] = 0
dt['-outFile'] = finalOut
dt['-o'][1] = finalOut + "/" + defaultLambdaFile + str(0)
try:
os.mkdir(dt['-o'][1])
except:
print("ERROR: Cannot create directory " + d['-o'][1])
exit(2)
os.system("cp " + finalOut + "/" + tempLabelsFile + " " + dt['-o'][1] + "/")
m, cvals = learnDiffLambda(dt, outfile, count, ds, trainSets, testSets)
sf.saveDetails(m, dt['-o'][1] + "/", d['-i'], cvals, dt['-tss'], 0, dt['-plotExtra'], dt['-pCol'], dt['-sortBy'], dt['-eps'])
bestCVL = best[1]
i = 1
while(1):
print "\n\nTrying Lambda", i, "\n\n"
dt['-lambda'] = i
dt['-o'][1] = finalOut + "/" + defaultLambdaFile + str(i)
try:
os.mkdir(dt['-o'][1])
except:
print("ERROR: Cannot create directory " + d['-o'][1])
exit(2)
os.system("cp " + finalOut + "/" + tempLabelsFile + " " + dt['-o'][1] + "/")
m1, cvals1 = learnDiffLambda(dt, outfile, count, ds, trainSets, testSets)
sf.saveDetails(m1, dt['-o'][1] + "/", d['-i'], cvals1, dt['-tss'], 0, dt['-plotExtra'], dt['-pCol'], dt['-sortBy'], dt['-eps'])
posMin = min(m1['m']['posCount'])
if posMin == 0 or best[1] < bestCVL: # Exit when minimum number of important features is 0 or when best cross validation likelihood is lesser compared to the one by previous lambda
del m1, cvals1
break
bestCVL = best[1]
del m, cvals
m = m1
cvals = cvals1
if posMin < 5: break
i = i + 2
for i in range(dt['-kfold']):
libctest.freeData(trainSets[i])
libctest.freeData(testSets[i])
libctest.freeData(ds)
del trainSets, testSets
return m, cvals
def learn(dt, outfile, count):
global d, ds, trainSets, testSets, lcount
lcount = dt['-lcount']
ds = getData(libctest.getData(dt['-f'], outfile))
pos = libctest.posList(ds.contents.n)
# Learn best model directly if -minarch and -maxarch are same
# if dt['-maxarch'] == dt['-minarch']:
# d = dt
# printBestModel(dt['-maxarch'])
# m = learnModel(dt['-maxarch'], dt['-o'][1] + "/" + tempFile)
# return m, []
for i in range(dt['-kfold']): # Get randomized train sets and test sets for every fold
trainSets.append(getData(libctest.getTrainSubset(ds, i, dt['-kfold'], pos)))
testSets.append(getData(libctest.getTestSubset(ds, i, dt['-kfold'], pos)))
if dt['-lambda'] != -1: # Learn models by single lambda value
m, cvals = learnDiffLambda(dt, outfile, count, ds, trainSets, testSets)
else: # Learn models by varying lambda
print "\n\nTrying Lambda", 0, "\n\n"
finalOut = dt['-o'][1]
dt['-lambda'] = 0
dt['-outFile'] = finalOut
dt['-o'][1] = finalOut + "/" + defaultLambdaFile + str(0)
try:
os.mkdir(dt['-o'][1])
except:
print("ERROR: Cannot create directory " + d['-o'][1])
exit(2)
os.system("cp " + finalOut + "/" + tempLabelsFile + " " + dt['-o'][1] + "/")
m, cvals = learnDiffLambda(dt, outfile, count, ds, trainSets, testSets)
sf.saveDetails(m, dt['-o'][1] + "/", d['-i'], cvals, dt['-tss'], 0, dt['-plotExtra'], dt['-pCol'], dt['-sortBy'], dt['-eps'])
bestCVL = best[1]
i = 1
while(1):
print "\n\nTrying Lambda", i, "\n\n"
dt['-lambda'] = i
dt['-o'][1] = finalOut + "/" + defaultLambdaFile + str(i)
try:
os.mkdir(dt['-o'][1])
except:
print("ERROR: Cannot create directory " + d['-o'][1])
exit(2)
os.system("cp " + finalOut + "/" + tempLabelsFile + " " + dt['-o'][1] + "/")
m1, cvals1 = learnDiffLambda(dt, outfile, count, ds, trainSets, testSets)
sf.saveDetails(m1, dt['-o'][1] + "/", d['-i'], cvals1, dt['-tss'], 0, dt['-plotExtra'], dt['-pCol'], dt['-sortBy'], dt['-eps'])
posMin = min(m1['m']['posCount'])
if posMin == 0 or best[1] < bestCVL: # Exit when minimum number of important features is 0 or when best cross validation likelihood is lesser compared to the one by previous lambda
del m1, cvals1
break
bestCVL = best[1]
del m, cvals
m = m1
cvals = cvals1
if posMin < 5: break
i = i + 2
for i in range(dt['-kfold']):
libctest.freeData(trainSets[i])
libctest.freeData(testSets[i])
libctest.freeData(ds)
del trainSets, testSets
return m, cvals
