def copy(self, shallow=False):
dmat = Sparse3DMatrix.copy(self)
dmat.count = self.count.copy()
dmat.lengths = self.lengths.copy()
dmat.num_loci, dmat.num_haplotypes, dmat.num_reads = dmat.shape
if not shallow:
dmat.__copy_names(self)
dmat.__copy_group_info(self)
return dmat
def __mul__(self, other):
dmat = Sparse3DMatrix.__mul__(self, other)
dmat.num_loci, dmat.num_haplotypes, dmat.num_reads = dmat.shape
if isinstance(
other,
(np.ndarray, csc_matrix, csr_matrix, coo_matrix, lil_matrix)):
dmat.hname = self.hname
dmat.rname = copy.copy(self.rname)
dmat.rid = copy.copy(self.rid)
dmat.num_groups = 0
else:
dmat.__copy_names(self)
dmat.__copy_group_info(self)
return dmat
def combine(self, other, shallow=False):
if self.finalized and other.finalized:
dmat = Sparse3DMatrix.combine(self, other)
dmat.num_loci, dmat.num_haplotypes, dmat.num_reads = dmat.shape
if self.count is not None and other.count is not None:
dmat.count = np.concatenate((self.count, other.count))
if self.lengths is not None:
dmat.lengths = copy.copy(self.lengths)
if not shallow:
dmat.hname = self.hname
dmat.lname = copy.copy(self.lname)
dmat.rname = np.concatenate((self.rname, other.rname))
dmat.lid = copy.copy(self.lid)
dmat.rid = dict(zip(dmat.rname, np.arange(dmat.num_reads)))
dmat.__copy_group_info(self)
return dmat
else:
raise RuntimeError('Both matrices must be finalized.')
def bundle(
self,
reset=False,
shallow=False): # Copies the original matrix (Use lots of memory)
"""
Returns ``AlignmentPropertyMatrix`` object in which loci are bundled using grouping information.
:param reset: whether to reset the values at the loci
:param shallow: whether to copy all the meta data
"""
if self.finalized:
# if self.num_groups > 0:
if self.groups is not None and self.gname is not None:
grp_conv_mat = lil_matrix((self.num_loci, self.num_groups))
for i in xrange(self.num_groups):
grp_conv_mat[self.groups[i], i] = 1.0
grp_align = Sparse3DMatrix.__mul__(
self, grp_conv_mat) # The core of the bundling
grp_align.num_loci = self.num_groups
grp_align.num_haplotypes = self.num_haplotypes
grp_align.num_reads = self.num_reads
grp_align.shape = (grp_align.num_loci,
grp_align.num_haplotypes,
grp_align.num_reads)
grp_align.count = self.count
if not shallow:
grp_align.lname = copy.copy(self.gname)
grp_align.hname = self.hname
grp_align.rname = copy.copy(self.rname)
grp_align.sname = copy.copy(self.sname)
grp_align.lid = dict(
zip(grp_align.lname, np.arange(grp_align.num_loci)))
grp_align.rid = copy.copy(self.rid)
grp_align.sid = copy.copy(self.sid)
if reset:
grp_align.reset()
return grp_align
else:
raise RuntimeError(
'No group information is available for bundling.')
else:
raise RuntimeError('The matrix is not finalized.')
def save(self,
h5file,
title=None,
index_dtype='uint32',
data_dtype=float,
incidence_only=True,
complib='zlib',
shallow=False):
Sparse3DMatrix.save(self,
h5file=h5file,
title=title,
index_dtype=index_dtype,
data_dtype=data_dtype,
incidence_only=incidence_only,
complib=complib)
h5fh = tables.open_file(h5file, 'a')
fil = tables.Filters(complevel=1, complib=complib)
if self.lengths is not None:
h5fh.create_carray(h5fh.root,
'lengths',
obj=self.lengths,
title='Transcript Lengths',
filters=fil)
if self.count is not None:
if len(self.count.shape) == 1: # count is a vector
h5fh.create_carray(h5fh.root,
'count',
obj=self.count,
title='Equivalence Class Counts',
filters=fil)
elif len(self.count.shape) == 2: # count is 2-dim matrix
if not isinstance(self.count, csc_matrix):
self.count = csc_matrix(self.count)
self.count.eliminate_zeros()
cgroup = h5fh.create_group(
h5fh.root, 'count',
'Sparse matrix components for N matrix')
h5fh.create_carray(cgroup,
'indptr',
obj=self.count.indptr.astype(index_dtype),
filters=fil)
h5fh.create_carray(cgroup,
'indices',
obj=self.count.indices.astype(index_dtype),
filters=fil)
h5fh.create_carray(cgroup,
'data',
obj=self.count.data.astype(index_dtype),
filters=fil)
if not shallow:
h5fh.set_node_attr(h5fh.root, 'hname', self.hname)
h5fh.create_carray(h5fh.root,
'lname',
obj=self.lname,
title='Locus Names',
filters=fil)
if self.rname is not None:
h5fh.create_carray(h5fh.root,
'rname',
obj=self.rname,
title='Read Names',
filters=fil)
if self.sname is not None:
h5fh.create_carray(h5fh.root,
'sname',
obj=self.sname,
title='Sample Names',
filters=fil)
h5fh.flush()
h5fh.close()
def __sub__(self, other):
dmat = Sparse3DMatrix.__sub__(self, other)
dmat.num_loci, dmat.num_haplotypes, dmat.num_reads = self.shape
dmat.__copy_names(self)
dmat.__copy_group_info(self)
return dmat
def __init__(self,
other=None,
ecfile=None,
h5file=None,
datanode='/',
metanode='/',
shallow=False,
shape=None,
dtype=float,
haplotype_names=None,
locus_names=None,
read_names=None,
sample_names=None,
grpfile=None):
Sparse3DMatrix.__init__(self,
other=other,
h5file=h5file,
datanode=datanode,
shape=shape,
dtype=dtype)
self.num_loci, self.num_haplotypes, self.num_reads = self.shape
self.num_samples = 0
self.num_groups = 0
self.count = None
self.hname = None
self.lname = None # locus name
self.rname = None # read name
self.sname = None # sample name (e.g. sample barcodes::cell barcodes)
self.lid = None # locus ID
self.rid = None # read ID
self.sid = None # sample ID
self.lengths = None # transcript lengths (or effective lengths)
self.gname = None # group name
self.groups = None # groups in terms of locus IDs
if other is not None: # Use for copying from other existing AlignmentPropertyMatrix object
if other.count is not None:
self.count = copy.copy(other.count)
if other.lengths is not None:
self.lengths = copy.copy(other.lengths)
if not shallow:
self.__copy_names(other)
self.__copy_group_info(other)
elif ecfile is not None:
with open(ecfile, 'rb') as f:
ecformat = unpack('<i', f.read(4))[0]
if ecformat == 2:
self.num_haplotypes = unpack('<i', f.read(4))[0]
hname = list()
for hidx in range(self.num_haplotypes):
hname_len = unpack('<i', f.read(4))[0]
hname.append(
unpack('<{}s'.format(hname_len),
f.read(hname_len))[0].decode('utf-8'))
self.hname = np.array(hname)
# Get transcipt info
self.num_loci = unpack('<i', f.read(4))[0]
self.lengths = np.zeros(
(self.num_loci, self.num_haplotypes), dtype=float)
tname = list()
for tidx in range(self.num_loci):
tname_len = unpack('<i', f.read(4))[0]
tname.append(
unpack('<{}s'.format(tname_len),
f.read(tname_len))[0].decode('utf-8'))
for hidx in range(self.num_haplotypes):
self.lengths[tidx, hidx] = unpack('<i',
f.read(4))[0]
self.lname = np.array(tname)
self.lid = dict(zip(self.lname, np.arange(self.num_loci)))
# Get sample info
sname = list()
self.num_samples = unpack('<i', f.read(4))[0]
for sidx in range(self.num_samples):
sname_len = unpack('<i', f.read(4))[0]
sname.append(
unpack('<{}s'.format(sname_len),
f.read(sname_len))[0].decode('utf-8'))
self.sname = np.array(sname)
self.sid = dict(
zip(self.sname, np.arange(self.num_samples)))
# Read in alignment matrix info
indptr_len = unpack('<i', f.read(4))[0]
self.num_reads = indptr_len - 1
nnz = unpack('<i', f.read(4))[0]
indptr_A = np.array(
unpack('<{}i'.format(indptr_len),
f.read(4 * indptr_len)))
indices_A = np.array(
unpack('<{}i'.format(nnz), f.read(4 * nnz)))
data_A = np.array(
unpack('<{}i'.format(nnz), f.read(4 * nnz)))
# Read in EC count matrix info
indptr_len = unpack('<i', f.read(4))[0]
nnz = unpack('<i', f.read(4))[0]
indptr_N = np.array(
unpack('<{}i'.format(indptr_len),
f.read(4 * indptr_len)))
indices_N = np.array(
unpack('<{}i'.format(nnz), f.read(4 * nnz)))
data_N = np.array(
unpack('<{}i'.format(nnz), f.read(4 * nnz)))
# Populate class member variables
for hidx in range(self.num_haplotypes - 1):
data_A, data_A_rem = np.divmod(data_A, 2)
self.data.append(
csr_matrix((data_A_rem, indices_A, indptr_A),
shape=(self.num_reads, self.num_loci)))
self.data.append(
csr_matrix((data_A, indices_A, indptr_A),
shape=(self.num_reads, self.num_loci)))
for hidx in range(self.num_haplotypes):
self.data[hidx].eliminate_zeros()
self.count = csc_matrix(
(data_N, indices_N, indptr_N),
shape=(self.num_reads, self.num_samples))
if self.num_samples == 1:
self.count = self.count.todense().A.flatten()
self.shape = (self.num_loci, self.num_haplotypes,
self.num_reads)
self.finalize()
elif ecformat == 1:
raise NotImplementedError
elif ecformat == 0:
raise TypeError('Format 0 is not supported anymore.')
elif h5file is not None: # Use for loading from a pytables file
h5fh = tables.open_file(h5file, 'r')
if not shallow:
self.hname = h5fh.get_node_attr(datanode, 'hname')
self.lname = np.char.decode(
h5fh.get_node(metanode, 'lname').read(), 'utf-8')
self.lid = dict(zip(self.lname, np.arange(self.num_loci)))
if h5fh.__contains__('%s' % (metanode + '/rname')):
self.rname = np.char.decode(
h5fh.get_node(metanode, 'rname').read(), 'utf-8')
self.rid = dict(zip(self.rname, np.arange(self.num_reads)))
if h5fh.__contains__('%s' % (metanode + '/sname')):
self.sname = np.char.decode(
h5fh.get_node(metanode, 'sname').read(), 'utf-8')
self.sid = dict(
zip(self.sname, np.arange(self.num_samples)))
self.num_samples = len(self.sname)
if h5fh.__contains__('%s' % (datanode + '/count')):
try:
self.count = h5fh.get_node(datanode,
'count').read() # Format-1
except tables.NoSuchNodeError as e: # Format-2
nmat_node = h5fh.get_node(datanode + '/count')
indptr = h5fh.get_node(nmat_node, 'indptr').read()
indices = h5fh.get_node(nmat_node, 'indices').read()
data = h5fh.get_node(nmat_node, 'data').read()
self.count = csc_matrix((data, indices, indptr),
dtype=np.float64)
self.num_samples = self.count.shape[1]
if h5fh.__contains__('%s' % (datanode + '/lengths')):
self.lengths = h5fh.get_node(datanode, 'lengths').read()
h5fh.close()
elif shape is not None: # Use for initializing an empty matrix
if haplotype_names is not None:
if len(haplotype_names) == self.num_haplotypes:
self.hname = haplotype_names
else:
raise RuntimeError(
'The number of names does not match to the matrix shape.'
)
if locus_names is not None:
if len(locus_names) == self.num_loci:
self.lname = np.array(locus_names)
self.lid = dict(zip(self.lname, np.arange(self.num_loci)))
else:
raise RuntimeError(
'The number of names does not match to the matrix shape.'
)
if read_names is not None:
if len(read_names) == self.num_reads:
self.rname = np.array(read_names)
self.rid = dict(zip(self.rname, np.arange(self.num_reads)))
else:
raise RuntimeError(
'The number of names does not match to the matrix shape.'
)
if sample_names is not None:
self.sname = np.array(sample_names)
self.sid = dict(zip(self.sname, np.arange(self.num_samples)))
self.num_samples = len(sample_names)
else:
self.num_samples = 1
if grpfile is not None:
self.__load_groups(grpfile)