def tspim2tensorformat(tsfile,
tensorfile,
tunit='s',
tbins='h',
idstartzero=True):
offset = 0 if idstartzero else -1
propdict = {}
with myreadfile(tsfile, 'rb') as fts, myreadfile(tensorfile, 'wb') as fte:
for line in fts:
k, v = line.strip().split(':')
propdict[k] = [v]
for k, vs in propdict.iteritems():
u, b = k.strip().split('-')
if idstartzero is False:
u = str(int(u) + offset)
b = str(int(b) + offset)
tss = vs[0].strip().split(' ')
tss = map(int, tss)
if tunit == 's':
'time unit is second'
if tbins == 'h':
'time bin size is hour'
tss = np.array(tss, dtype=int) / 3600
elif tbins == 'd':
'time bin size is day'
tss = np.array(tss, dtype=int) / (3600 * 24)
if type(tbins) is int:
tss = np.array(tss, dtype=int) / tbins
tss = map(str, tss)
for i in xrange(len(tss)):
fte.write(','.join((u, b, tss[i], '1')))
fte.write('\n')
fts.close()
fte.close()
return
def load_from_edgeproperty(self, profnm, mtype=coo_matrix, dtype=int):
'load the graph edge property, time stamps, ratings, or text vector'
self.idstartzero = True #record for output recovery
offset = -1 if self.idstartzero is False else 0
'sparse matrix has special meaning of 0, so property index start from 1'
self.eprop = [np.array([])
] #make the idx start from 1 in sparse matrix
with myreadfile(profnm, 'rb') as fin:
idx = 1
x, y, data = [], [], []
for line in fin:
um, prop = line.strip().split(':')
u, m = um.split('-')
u = int(u) + offset
m = int(m) + offset
x.append(u)
y.append(m)
data.append(idx) #data store the index of edge properties
prop = np.array(prop.strip().split()).astype(dtype)
self.eprop.append(prop)
idx += 1
fin.close()
self.edgeidxm = mtype((data, (x, y)),
shape=(max(x) + 1, max(y) + 1))
self.edgeidxmr = self.edgeidxm.tocsr()
self.edgeidxmc = self.edgeidxm.tocsc()
self.edgeidxml = self.edgeidxm.tolil()
self.edgeidxmlt = self.edgeidxm.transpose().tolil()
self.eprop = np.array(self.eprop)
return
def loadtensor2matricization(tensorfile, sumout=[], mtype=coo_matrix,
weighted=True, dtype=int):
'sumout: marginized (sumout) the given ways'
matcols={}
rindexcols={}
xs, ys, data = [], [], []
with myreadfile(tensorfile, 'rb') as f:
for line in f:
elems = line.strip().split(',')
elems = np.array(elems)
u = int(elems[0])
colidx = range(1,len(elems)-1) #remove sumout
colidx = set(colidx) - set(list(sumout))
colidx = sorted(list(colidx))
col=' '.join(elems[colidx])
if col not in matcols:
idx = len(matcols)
matcols[col] = idx
rindexcols[idx]=col
cid = matcols[col]
w = dtype(elems[-1])
xs.append(u)
ys.append(cid)
data.append(w)
nrow, ncol = max(xs)+1, max(ys)+1
sm = mtype( (data, (xs, ys)), shape=(nrow, ncol), dtype=dtype )
if weighted is False:
sm.data[0:] = dtype(1)
f.close()
return sm, rindexcols
def pim2tensorformat(tsfile, ratefile, tensorfile, tunit='s', tbins='h'):
'convert the pim files: tsfile, ratefile into tensor file, i.e. tuples'
rbins = lambda x: 0 if x < 2.5 else 1 if x <= 3.5 else 2 #lambda x: x
propdict = {}
with myreadfile(tsfile, 'rb') as fts, myreadfile(ratefile, 'rb') as frt,\
open(tensorfile, 'wb') as fte:
for line in fts:
k, v = line.strip().split(':')
propdict[k] = [v]
for line in frt:
k, v = line.strip().split(':')
propdict[k].append(v)
for k, vs in propdict.iteritems():
u, b = k.strip().split('-')
tss = vs[0].strip().split(' ')
tss = map(int, tss)
if tunit == 's':
'time unit is second'
if tbins == 'h':
'time bin size is hour'
tss = np.array(tss, dtype=int) / 3600
elif tbins == 'd':
'time bin size is day'
tss = np.array(tss, dtype=int) / (3600 * 24)
'no matter what the tunit is'
if type(tbins) is int:
tss = np.array(tss, dtype=int) / tbins
tss = map(str, tss)
'process ts'
rts = vs[1].strip().split(' ')
rts = map(float, rts)
digrs = []
for r1 in rts:
r = rbins(r1)
digrs.append(r)
digrs = map(int, digrs)
digrs = map(str, digrs)
for i in xrange(len(tss)):
fte.write(','.join((u, b, tss[i], digrs[i], '1')))
fte.write('\n')
fts.close()
frt.close()
fte.close()
return
