def generate_dense_cpt_objects(self):
# names of dense cpts
names = [
"start_seg", "seg_subseg", "seg_seg", "seg_subseg_subseg",
"segCountDown_seg_segTransition"
]
num_segs = self.num_segs
num_subsegs = self.num_subsegs
# create required probability tables
start_seg = fill_array(1.0 / num_segs, num_segs)
seg_subseg = fill_array(1.0 / num_subsegs, (num_segs, num_subsegs))
seg_seg = make_zero_diagonal_table(num_segs)
cpt_seg = make_zero_diagonal_table(num_subsegs)
seg_subseg_subseg = (vstack_tile(cpt_seg, num_segs, 1))
segCountDown = self.make_dense_cpt_segCountDown_seg_segTransition()
prob = [
start_seg, seg_subseg, seg_seg, seg_subseg_subseg, segCountDown
]
# create DenseCPTs and add to input_master.dense_cpt: InlineSection
for i in range(len(names)):
input_master.dense_cpt[names[i]] = np.squeeze(DenseCPT(prob[i]),
axis=0)
# adding dirichlet row if necessary
if self.len_seg_strength > 0:
dirichlet_row = [
"DirichletTable %s" %
self.make_dirichlet_name(NAME_SEGCOUNTDOWN_SEG_SEGTRANSITION)
]
input_master.dense_cpt[
NAME_SEGCOUNTDOWN_SEG_SEGTRANSITION].extra_rows = dirichlet_row
return str(input_master.dense_cpt)
def make_seg_subseg(self):
name = "seg_subseg"
parent_card = self.num_segs # TODO check
card = self.num_subseg # TODO check
prob = fill_array(1.0 / self.num_subsegs, (self.num_segs,
self.num_subsegs))
seg_subseg = DenseCPT(name=name, parent_card=parent_card,
cardinality=card, prob=prob)
input_master.update(seg_subseg)
def make_zero_diagonal_table(length):
if length == 1:
return array([1.0]) # always return to self
prob_self_self = 0.0
prob_self_other = (1.0 - prob_self_self) / (length - 1)
# set everywhere (diagonal to be rewritten)
res = fill_array(prob_self_other, (length, length))
# set diagonal
range_cpt = range(length)
res[range_cpt, range_cpt] = prob_self_self
return res
def write_metadata(chromosome, verbose=False):
if verbose:
print("writing metadata for %s" % chromosome, file=sys.stderr)
if chromosome.num_tracks_continuous == 0:
chromosome.attrs.dirty = False
return
tracknames = chromosome.tracknames_continuous
num_obs = len(tracknames)
row_shape = (num_obs, )
mins = fill_array(PINF, row_shape)
maxs = fill_array(NINF, row_shape)
sums = fill_array(0.0, row_shape)
sums_squares = fill_array(0.0, row_shape)
num_datapoints = fill_array(0, row_shape)
supercontigs = chromosome.supercontigs[chromosome.start:chromosome.end]
num_supercontigs = len(supercontigs)
prev_chrom_end = 0 # keeps track of where data was last present
prev_supercontigs = [None] + supercontigs[:-1]
next_supercontigs = supercontigs[1:] + [None]
zipper = zip(prev_supercontigs, supercontigs, next_supercontigs)
for prev_supercontig, supercontig, next_supercontig in zipper:
if verbose:
print(" scanning %s" % supercontig, file=sys.stderr)
try:
continuous = supercontig.continuous
except NoSuchNodeError:
raise NoSuchNodeError("Supercontig found missing continuous")
# only runs when assertions checked
if __debug__:
init_num_obs(num_obs, continuous) # for the assertion
# An array containing True when any data is present across all tracks
mask_rows_any_present = fill_array(False, continuous.shape[0])
# doing this column by column greatly reduces the memory
# footprint when you have large numbers of tracks. It also
# simplifies the logic for the summary stats, since you don't
# have to change the mask value for every operation, like in
# revisions <= r243
for col_index, trackname in enumerate(tracknames):
if verbose:
print(" %s" % trackname, file=sys.stderr)
## read data
col = continuous[:, col_index]
mask_present = isfinite(col)
mask_rows_any_present[mask_present] = True
col_finite = col[mask_present]
del col # col not needed anymore (optimization)
num_datapoints_col = len(col_finite)
if num_datapoints_col:
update_extrema(get_nonzero_min, mins, col_finite, col_index)
# print(update_extrema(get_nonzero_min, mins, col_finite, col_index))
update_extrema(amax, maxs, col_finite, col_index)
sums[col_index] += col_finite.sum(0)
sums_squares[col_index] += square(col_finite).sum(0)
num_datapoints[col_index] += num_datapoints_col
supercontig_attrs = supercontig.attrs
next_chrom_start = find_next_present_coord(chromosome,
next_supercontig)
# If any data is present in this supercontig
if any(mask_rows_any_present):
# Find all gaps between present data
starts, ends = find_chunk_gaps_in_supercontig(
mask_rows_any_present)
# Trim the starting chunk index for this supercontig
starts[0] = trim_chunks_start(prev_supercontig, prev_chrom_end,
supercontig, mask_rows_any_present,
starts[0])
# Trim the end chunk index for this supercontig
ends[-1] = trim_chunks_end(next_supercontig, next_chrom_start,
supercontig, mask_rows_any_present,
ends[-1])
# Update our new previously present coordinate for next iteration
prev_chrom_end = supercontig.start + ends[-1]
# Otherwise there isn't any data present for this contig
else:
# If this is the last supercontig
if not next_supercontig:
# Exclude the entire region
starts = array([])
ends = array([])
# Otherwise there are supercontigs to lookahead at
else:
# If the distance between the previously last present value
# And next possible present value is less than MIN_GAP_LEN
if (next_chrom_start
and next_chrom_start - prev_chrom_end < MIN_GAP_LEN):
# Include the entire region
starts = array([0])
ends = array([mask_rows_any_present.shape[0]])
# Otherwise
else:
# Exclude the entire region
starts = array([])
ends = array([])
supercontig_attrs.chunk_starts = starts
supercontig_attrs.chunk_ends = ends
chromosome_attrs = chromosome.attrs
chromosome_attrs.mins = mins
chromosome_attrs.maxs = maxs
chromosome_attrs.sums = sums
chromosome_attrs.sums_squares = sums_squares
chromosome_attrs.num_datapoints = num_datapoints
chromosome_attrs.dirty = False
tupler = (mins, maxs)
# print(tupler)
return (mins, maxs)
def make_dense_cpt_seg_subseg_spec(self):
num_subsegs = self.num_subsegs
cpt = fill_array(1.0 / num_subsegs, (self.num_segs, num_subsegs))
return self.make_table_spec("seg_subseg", cpt)
def make_start_seg(self):
name = "start_seg"
card = self.num_segs
prob = fill_array(1.0 / self.num_segs, self.num_segs)
start_seg = DenseCPT(name=name, cardinality=card, prob=prob)
input_master.update(start_seg)