def outside_grow(spec, min_radius, i_merge_bblock, **kw):
bbspec0, crit0 = spec
bbspec = deepcopy(bbspec0[crit0.from_seg :])
bbspec[0][1] = "_" + bbspec[0][1][1]
print("outside", bbspec)
if os.path.exists("outer.pickle"):
with open("outer.pickle", "rb") as inp:
ssdag, rslt, crit = pickle.load(inp)
else:
crit = Cyclic(crit0.nfold, min_radius=min_radius)
ssdag = simple_search_dag(
bbspec, modbbs=modsinglebb((0, -1), i_merge_bblock), **kw
)
rslt = grow_linear(
ssdag,
loss_function=crit.jit_lossfunc(),
loss_threshold=1.0,
last_bb_same_as=crit.from_seg,
**kw,
)
with open("outer.pickle", "wb") as out:
pickle.dump((ssdag, rslt, crit), out)
return ssdag, rslt, crit
def inside_grow(spec, binner, table, i_merge_bblock, **kw):
bbspec0, crit0 = spec
bbspec = deepcopy(bbspec0[:crit0.from_seg + 1])
bbspec[-1][1] = bbspec[-1][1][0] + '_'
print('inside', bbspec)
if os.path.exists('inner.pickle'):
with open('inner.pickle', 'rb') as inp:
ssdag, rslt = pickle.load(inp)
else:
ssdag = simple_search_dag(bbspec, modbbs=modsinglebb((-1, ), i_merge_bblock),
**kw)
rslt = grow_linear(
ssdag,
loss_function=_hash_lossfunc(binner, table, crit0.nfold),
# loss_function=lossfunc_rand_1_in(1000),
loss_threshold=1.0,
**kw
)
with open('inner.pickle', 'wb') as out:
pickle.dump((ssdag, rslt), out)
return ssdag, rslt
def worm_grow_3(bbdb,
spdb,
nbblocks=10,
shuffle_bblocks=0,
parallel=1,
verbosity=1,
monte_carlo=0,
clash_check=0,
dump_pdb=0,
cache_sync=0.001):
if clash_check < dump_pdb: clash_check = dump_pdb * 100
ttot = time()
ssdag, tdb, tvertex, tedge = simple_search_dag(
[
('C3_N', '_N'),
('Het:NCy', 'C_'),
# ('Het:CCC', 'C_'),
# ('Het:NN', 'NN'),
# ('Het:CC', 'CC'),
# ('Het:NNX', 'N_'),
],
(bbdb, spdb),
nbblocks=nbblocks,
timing=True,
verbosity=verbosity,
parallel=parallel,
cache_sync=cache_sync)
# crit = Cyclic(3, from_seg=2, origin_seg=0)
# crit = Cyclic(3)
# last_bb_same_as = crit.from_seg
crit = NullCriteria()
lf = crit.jit_lossfunc()
last_bb_same_as = -1
tgrow = time()
rslt = grow_linear(
ssdag,
# loss_function=lf,
loss_function=lossfunc_rand_1_in(1000),
parallel=parallel,
loss_threshold=1.0,
last_bb_same_as=last_bb_same_as,
monte_carlo=monte_carlo)
tgrow = time() - tgrow
Nres = len(rslt.err)
Ntot = np.prod([v.len for v in ssdag.verts])
logtot = np.log10(Ntot)
print('frac last_bb_same_as',
rslt.stats.n_last_bb_same_as[0] / rslt.stats.total_samples[0])
Nsparse = int(rslt.stats.total_samples[0])
Nsparse_rate = int(Nsparse / tgrow)
ttot = time() - ttot
if len(rslt.idx) == 0: frac_redundant = 0
else: frac_redundant = rslt.stats.n_redundant_results[0] / len(rslt.idx)
print(
f' worm_grow_3 {nbblocks:4} {ttot:7.1f}s {Nres:9,} logtot{logtot:4.1f} tv'
f' {tvertex:7.1f}s te {tedge:7.1f}s tg {tgrow:7.1f}s {Nsparse:10,} {Nsparse_rate:7,}/s {frac_redundant:4.1f}'
)
if len(rslt.err):
print('err 0 25 50 75 100',
np.percentile(rslt.err, (0, 25, 50, 75, 100)))
sys.stdout.flush()
if not clash_check: return
tclash = time()
norig = len(rslt.idx)
# rslt = prune_clashes(
# ssdag, crit, rslt, at_most=clash_check, thresh=4.0, parallel=parallel
# )
print(
'pruned clashes, %i of %i remain,' %
(len(rslt.idx), min(clash_check, norig)), 'took',
time() - tclash, 'seconds')
for i, idx in enumerate(rslt.idx[:10]):
graph_dump_pdb('graph_%i_nojoin.pdb' % i,
ssdag,
idx,
rslt.pos[i],
join=0)
# graph_dump_pdb('graph_%i.pdb' % i, ssdag, idx, rslt.pos[i])
return
if len(rslt.idx) > 0:
tpdb = time()
exe = cf.ThreadPoolExecutor if parallel else InProcessExecutor
with exe(max_workers=3) as pool:
futures = list()
for i in range(min(dump_pdb, len(rslt.idx))):
kw = dict(
bbdb=bbdb,
ssdag=ssdag,
# crit=crit,
i=i,
indices=rslt.idx[i],
positions=rslt.pos[i],
only_connected=False,
)
futures.append(pool.submit(_dump_pdb, **kw))
[f.result() for f in futures]
print('dumped %i structures' % min(dump_pdb, len(rslt.idx)), 'time',
time() - tpdb)