def arffread(kernelname,datafilename):
"""Decide based on kernelname whether to read a sequence or vectorial file"""
if kernelname == 'gauss' or kernelname == 'linear' or kernelname == 'poly' or kernelname == None:
fp = init_datasetfile(datafilename,'vec')
elif kernelname == 'wd' or kernelname == 'localalign' or kernelname == 'localimprove'\
or kernelname == 'spec' or kernelname == 'cumspec':
fp = init_datasetfile(datafilename,'seq')
elif kernelname == 'spec2' or kernelname == 'cumspec2':
fp = init_datasetfile(datafilename,'mseq')
else:
print 'Unknown kernel in arffread'
return fp.readlines()
def arffread(kernelname, datafilename):
"""Decide based on kernelname whether to read a sequence or vectorial file"""
if kernelname == 'gauss' or kernelname == 'linear' or kernelname == 'poly' or kernelname == None:
fp = init_datasetfile(datafilename, 'vec')
elif kernelname == 'wd' or kernelname == 'localalign' or kernelname == 'localimprove'\
or kernelname == 'spec' or kernelname == 'cumspec':
fp = init_datasetfile(datafilename, 'seq')
elif kernelname == 'spec2' or kernelname == 'cumspec2':
fp = init_datasetfile(datafilename, 'mseq')
else:
print 'Unknown kernel in arffread'
return fp.readlines()
def fastaread(fnamepos, fnameneg=None):
"""Read two fasta files, the first positive, the second negative"""
fpos = init_datasetfile(fnamepos, 'seq')
(fa1, lab1) = fpos.readlines()
if fnameneg is not None:
fneg = init_datasetfile(fnameneg, 'seq')
(fa2, lab2) = fneg.readlines()
print 'positive: %d, negative %d' % (len(fa1), len(fa2))
all_labels = concatenate((ones(len(fa1)), -ones(len(fa2))))
all_examples = fa1 + fa2
else:
all_examples = fa1
all_labels = ones(len(fa1))
return all_examples, all_labels
def fastaread(fnamepos,fnameneg=None):
"""Read two fasta files, the first positive, the second negative"""
fpos = init_datasetfile(fnamepos,'seq')
(fa1,lab1) = fpos.readlines()
if fnameneg is not None:
fneg = init_datasetfile(fnameneg,'seq')
(fa2,lab2) = fneg.readlines()
print 'positive: %d, negative %d' % (len(fa1),len(fa2))
all_labels = concatenate((ones(len(fa1)),-ones(len(fa2))))
all_examples = fa1 + fa2
else:
all_examples = fa1
all_labels = ones(len(fa1))
return all_examples, all_labels
def fastawrite_sequence(filename,p):
"""Write a FASTA file containing a sequence dataset"""
import arff
(metadata,seqlist) = motifgen(p.motif, p.numseq, p.seqlenmin, p.seqlenmax, p.posstart, p.posend, p.mutrate)
labels = ones(len(seqlist))
fp = init_datasetfile(filename,'seq')
fp.writelines(seqlist,labels)
def arffwrite_real(filename, numpoint, numfeat, fracpos=0.5, width=1.0):
"""Write an ARFF file containing a vectorial dataset"""
#import arff
(metadata, pointcloud, labels) = cloudgen(numpoint, numfeat, fracpos, width)
fp = init_datasetfile(filename,'vec')
fp.comment = metadata
fp.dataname = 'pointcloud'
fp.writelines(pointcloud,labels)
def arffwrite_sequence(filename,p, n):
"""Write an ARFF file containing a sequence dataset"""
#import arff
(metadatapos,seqlistpos) = motifgen(p.motif, p.numseq, p.seqlenmin, p.seqlenmax, p.posstart, p.posend, p.mutrate)
(metadataneg,seqlistneg) = motifgen(n.motif, n.numseq, n.seqlenmin, n.seqlenmax, n.posstart, n.posend, n.mutrate)
labels = concatenate((ones(len(seqlistpos)),-ones(len(seqlistneg))))
seqlist = seqlistpos + seqlistneg
fp = init_datasetfile(filename,'seq')
fp.comment = metadatapos+' '+metadataneg
fp.dataname = 'motif'
fp.writelines(seqlist,labels)
def test_gc(gcfilename):
"""
Check the gc content files for conflicting labels
"""
fp = init_datasetfile(gcfilename,'vec')
(examples,labels) = fp.readlines()
print '%d positive and %d negative examples' % (sum(labels>0.0),sum(labels<0.0))
distance = sqr_dist(numpy.matrix(examples),numpy.matrix(examples))
labdist = numpy.matrix(labels).T*numpy.matrix(labels)
#difflab = numpy.where(labdist.A<0,distance,numpy.matlib.ones((len(labels),len(labels))))
contracount = 0
for ix in xrange(len(labels)):
for iy in xrange(ix+1,len(labels)):
if labdist[ix,iy]<0 and distance[ix,iy]<0.01:
contracount += 1
print distance.shape, labdist.shape
#print '%d identical examples with opposing labels' %len(numpy.unique(numpy.where(difflab==0)[0]))
print '%d identical examples with opposing labels' % contracount
def test_gc(gcfilename):
"""
Check the gc content files for conflicting labels
"""
fp = init_datasetfile(gcfilename, 'vec')
(examples, labels) = fp.readlines()
print '%d positive and %d negative examples' % (sum(labels > 0.0),
sum(labels < 0.0))
distance = sqr_dist(numpy.matrix(examples), numpy.matrix(examples))
labdist = numpy.matrix(labels).T * numpy.matrix(labels)
#difflab = numpy.where(labdist.A<0,distance,numpy.matlib.ones((len(labels),len(labels))))
contracount = 0
for ix in xrange(len(labels)):
for iy in xrange(ix + 1, len(labels)):
if labdist[ix, iy] < 0 and distance[ix, iy] < 0.01:
contracount += 1
print distance.shape, labdist.shape
#print '%d identical examples with opposing labels' %len(numpy.unique(numpy.where(difflab==0)[0]))
print '%d identical examples with opposing labels' % contracount
(n.seqlenmin, n.seqlenmax) = esvm.parse.parse_range(sys.argv[10])
(n.posstart, n.posend) = esvm.parse.parse_range(sys.argv[11])
n.mutrate = float(sys.argv[12])
filename = sys.argv[13]
arffwrite_sequence(filename, p, n)
elif sys.argv[1] == 'cloud':
# generate a data cloud in ARFF format
numpoint = int(sys.argv[2])
numfeat = int(sys.argv[3])
fracpos = float(sys.argv[4])
width = float(sys.argv[5])
filename = sys.argv[6]
arffwrite_real(filename, numpoint, numfeat, fracpos, width)
if len(sys.argv) >= 8:
fp = init_datasetfile(filename, 'vec')
(examples, labels) = fp.readlines()
pointcloud = []
for ix in xrange(numpoint):
pointcloud.append(
array([labels[ix], examples[0, ix], examples[1, ix]]))
esvm.plots.plotcloud(pointcloud, sys.argv[7], 'Pointcloud')
#(examples,labels,metadata)=arffwrite_real(filename, numpoint, numfeat, fracpos, width)
#if len(sys.argv)>=8:
# plots.plotcloud(pointcloud,sys.argv[7],metadata)
else:
print 'Unknown option %s\n' % sys.argv[1]
def splice_example(Cs, gcfilename,seqfilename,seq2filename, plot=False):
"""
For the data files, apply the set of kernels
"""
# hyperparameters
num_fold_cv = 5
# The area under the receiver operating characteristic
results=[]
# Read datasets
# GC features
fp = init_datasetfile(gcfilename,'vec')
(gc_examples,gc_labels) = fp.readlines()
gc_examples = normalize(gc_examples, subtract_mean=True)
if plot:
from pylab import scatter,show
color=['b','r']
scatter(gc_examples[0,], gc_examples[1,], s=400*(gc_labels+2), c=''.join([ color[(int(i)+1)/2] for i in gc_labels]), alpha=0.1)
show()
# 2 sequence features
fp = init_datasetfile(seq2filename,'mseq')
(dna2_examples,dna2_labels) = fp.readlines()
# DNA sequences
fp = init_datasetfile(seqfilename,'seq')
(dna_examples,dna_labels) = fp.readlines()
#Define experiments to carry out
experiments=(
# Linear kernel on GC content
('linear', {'scale':1.0, 'name':'scale'}, (gc_examples, gc_labels)),
# Polynomial kernel on GC content
( 'poly', {'degree':3, 'name':'degree', 'inhomogene':True, 'normal':True}, (gc_examples, gc_labels)),
( 'poly', {'degree':5, 'name':'degree', 'inhomogene':True, 'normal':True}, (gc_examples, gc_labels)),
# Gaussian kernel on GC content
('gauss', {'width':100.0, 'name':'width'}, (gc_examples, gc_labels)),
('gauss', {'width':1.0, 'name':'width'}, (gc_examples, gc_labels)),
('gauss', {'width':0.01, 'name':'width'}, (gc_examples, gc_labels)),
# Spectrum kernel on 2 dna sequences
('spec2', {'degree':1, 'name':'degree'}, (dna2_examples, dna2_labels)),
('spec2', {'degree':3, 'name':'degree'}, (dna2_examples, dna2_labels)),
('spec2', {'degree':5, 'name':'degree'}, (dna2_examples, dna2_labels)),
# Cumulative Spectrum kernel on 2 dna sequences
('cumspec2', {'degree':1, 'name':'degree'}, (dna2_examples, dna2_labels)),
('cumspec2', {'degree':3, 'name':'degree'}, (dna2_examples, dna2_labels)),
('cumspec2', {'degree':5, 'name':'degree'}, (dna2_examples, dna2_labels)),
# Weighted degree kernel on dna sequences
('wd', {'degree':1,'shift':0, 'name':'degree'}, (dna_examples, dna_labels)),
('wd', {'degree':3,'shift':0, 'name':'degree'}, (dna_examples, dna_labels)),
('wd', {'degree':5,'shift':0, 'name':'degree'}, (dna_examples, dna_labels))
)
if Cs is None:
for C in (0.01, 0.1, 1, 2, 5, 10):
for e in experiments:
run_single_experiment(results, num_fold_cv, e[0], e[1], C, e[2][0], e[2][1])
else:
for i in xrange(len(experiments)):
e=experiments[i]
run_single_experiment(results, num_fold_cv, e[0], e[1], Cs[i], e[2][0], e[2][1])
return results
def splice_example(Cs, gcfilename, seqfilename, seq2filename, plot=False):
"""
For the data files, apply the set of kernels
"""
# hyperparameters
num_fold_cv = 5
# The area under the receiver operating characteristic
results = []
# Read datasets
# GC features
fp = init_datasetfile(gcfilename, 'vec')
(gc_examples, gc_labels) = fp.readlines()
gc_examples = normalize(gc_examples, subtract_mean=True)
if plot:
from pylab import scatter, show
color = ['b', 'r']
scatter(gc_examples[0, ],
gc_examples[1, ],
s=400 * (gc_labels + 2),
c=''.join([color[(int(i) + 1) / 2] for i in gc_labels]),
alpha=0.1)
show()
# 2 sequence features
fp = init_datasetfile(seq2filename, 'mseq')
(dna2_examples, dna2_labels) = fp.readlines()
# DNA sequences
fp = init_datasetfile(seqfilename, 'seq')
(dna_examples, dna_labels) = fp.readlines()
#Define experiments to carry out
experiments = (
# Linear kernel on GC content
('linear', {
'scale': 1.0,
'name': 'scale'
}, (gc_examples, gc_labels)),
# Polynomial kernel on GC content
('poly', {
'degree': 3,
'name': 'degree',
'inhomogene': True,
'normal': True
}, (gc_examples, gc_labels)),
('poly', {
'degree': 5,
'name': 'degree',
'inhomogene': True,
'normal': True
}, (gc_examples, gc_labels)),
# Gaussian kernel on GC content
('gauss', {
'width': 100.0,
'name': 'width'
}, (gc_examples, gc_labels)),
('gauss', {
'width': 1.0,
'name': 'width'
}, (gc_examples, gc_labels)),
('gauss', {
'width': 0.01,
'name': 'width'
}, (gc_examples, gc_labels)),
# Spectrum kernel on 2 dna sequences
('spec2', {
'degree': 1,
'name': 'degree'
}, (dna2_examples, dna2_labels)),
('spec2', {
'degree': 3,
'name': 'degree'
}, (dna2_examples, dna2_labels)),
('spec2', {
'degree': 5,
'name': 'degree'
}, (dna2_examples, dna2_labels)),
# Cumulative Spectrum kernel on 2 dna sequences
('cumspec2', {
'degree': 1,
'name': 'degree'
}, (dna2_examples, dna2_labels)),
('cumspec2', {
'degree': 3,
'name': 'degree'
}, (dna2_examples, dna2_labels)),
('cumspec2', {
'degree': 5,
'name': 'degree'
}, (dna2_examples, dna2_labels)),
# Weighted degree kernel on dna sequences
('wd', {
'degree': 1,
'shift': 0,
'name': 'degree'
}, (dna_examples, dna_labels)),
('wd', {
'degree': 3,
'shift': 0,
'name': 'degree'
}, (dna_examples, dna_labels)),
('wd', {
'degree': 5,
'shift': 0,
'name': 'degree'
}, (dna_examples, dna_labels)))
if Cs is None:
for C in (0.01, 0.1, 1, 2, 5, 10):
for e in experiments:
run_single_experiment(results, num_fold_cv, e[0], e[1], C,
e[2][0], e[2][1])
else:
for i in xrange(len(experiments)):
e = experiments[i]
run_single_experiment(results, num_fold_cv, e[0], e[1], Cs[i],
e[2][0], e[2][1])
return results
n.numseq = int(sys.argv[9])
(n.seqlenmin,n.seqlenmax) = esvm.parse.parse_range(sys.argv[10])
(n.posstart,n.posend) = esvm.parse.parse_range(sys.argv[11])
n.mutrate = float(sys.argv[12])
filename = sys.argv[13]
arffwrite_sequence(filename, p, n)
elif sys.argv[1] == 'cloud':
# generate a data cloud in ARFF format
numpoint = int(sys.argv[2])
numfeat = int(sys.argv[3])
fracpos = float(sys.argv[4])
width = float(sys.argv[5])
filename = sys.argv[6]
arffwrite_real(filename, numpoint, numfeat, fracpos, width)
if len(sys.argv)>=8:
fp = init_datasetfile(filename,'vec')
(examples,labels) = fp.readlines()
pointcloud = []
for ix in xrange(numpoint):
pointcloud.append(array([labels[ix],examples[0,ix],examples[1,ix]]))
esvm.plots.plotcloud(pointcloud,sys.argv[7],'Pointcloud')
#(examples,labels,metadata)=arffwrite_real(filename, numpoint, numfeat, fracpos, width)
#if len(sys.argv)>=8:
# plots.plotcloud(pointcloud,sys.argv[7],metadata)
else:
print 'Unknown option %s\n' % sys.argv[1]
