extra = args.extra
root = args.root
nc = args.nc
target = args.target
ntwk = model(typem=typem, base=args.base, nc=args.nc)
ntwk.load_weights(weight_name)
if args.base:
root = '/data/bioinfo@borvo/users/jarbona/deepnano5bases/data/tomb/clean_name/'
X, y = load_data([filename], root=root, nc=args.nc)
if not args.base:
Xr, yr, fn = load_events(X,
y,
min_length=length_window,
base=args.base,
maxf=args.maxf)
extra_e = []
else:
Xr, yr, fn, extra_e = load_events(X,
y,
min_length=length_window,
base=args.base,
maxf=args.maxf,
extra=True)
assert (len(fn) == len(Xr))
print("Nfiles", len(Xr))
yr = np.array(yr)
Xt, yt, which_keep, NotT = transform_reads(Xr,
yr,
# print("la")
ws = 8
# print(ws)
X, y = load_data(
[t],
root=root,
values=[[
"test_with_tombo_CNV_logcosh_3layers/weights.22-0.01", 0
], [
"test_longueur_lstm_from_scratch_without_human_weights.25-0.02", 0
], ["init_B", 0], ["init_I", 1]],
nc=args.nc)
if not args.base:
Xrt, yrt, fnt = load_events(X,
y,
min_length=10 * length_window,
raw=args.raw,
base=args.base,
maxf=args.maxf)
extra_e = []
else:
Xrt, yrt, fnt, extra_e = load_events(X,
y,
min_length=10 * length_window,
raw=args.raw,
base=args.base,
maxf=args.maxf,
extra=True)
# print(Xrt[0])
if args.raw:
max_len = 10000
def load_data_complete(dataset,
root,
per_dataset=None,
lenv=200,
shuffle=True,
pmix=None,
values=[],
delta=False,
raw=False,
rescale=False,
base=False,
noise=False,
nc=1):
Tt = np.load("data/training/T-T1-corrected-transition_iter3.npy")
X_t, y_t = [], []
for data in dataset:
print("Loading", data)
ws = 5
if "T-yeast" in data:
ws = 8
t0 = time.time()
X, y = load_data([data],
root=root,
per_dataset=per_dataset,
values=values + [["init_B", 0], ["init_I", 1]],
nc=nc) # X filename,y B amount
t1 = time.time()
print(t1 - t0, "load csv")
t0 = time.time()
# X events y B amount filtered for length < 10000
if base:
Xp, yp, fn, extra_e = load_events(X,
y,
min_length=None,
ws=ws,
raw=raw,
base=base,
extra=True)
else:
extra_e = []
Xp, yp, fn = load_events(X,
y,
min_length=None,
ws=ws,
raw=raw,
base=base)
t1 = time.time()
print(t1 - t0, "load events")
t0 = time.time()
print("Mean Values", np.mean(yp, axis=0))
print("Total cumulated read length",
np.sum([len(xi["mean"]) for xi in Xp]))
assert (len(Xp) == len(yp))
Xpp, ypp, _, _ = transform_reads(Xp,
np.array(yp),
lenv=lenv,
delta=delta,
rescale=rescale,
noise=noise,
extra_e=extra_e,
Tt=Tt)
t1 = time.time()
print(t1 - t0, "transform")
t0 = time.time()
Xpp = np.concatenate(Xpp, axis=0)
ypp = np.concatenate(ypp, axis=0)
t1 = time.time()
print(t1 - t0, "concat")
t0 = time.time()
print("Total cumulated read length_after_cut", Xpp.shape[0])
X_t.append(Xpp)
y_t.append(ypp)
# print(X_t)
X_t = np.concatenate(X_t, axis=0)
y_t = np.concatenate(y_t, axis=0)
if pmix is not None:
print("Mixing", pmix)
a = np.arange(len(X_t))
m1 = np.random.choice(a, int(len(a) * pmix))
m2 = np.random.choice(a, int(len(a) * pmix))
nX = np.concatenate((X_t[m1, :100, ::], X_t[m2, 100:, ::]), axis=1)
ny = y_t[m1] / 2 + y_t[m2] / 2
X_t = np.concatenate([X_t, nX], axis=0)
y_t = np.concatenate([y_t, ny], axis=0)
if shuffle:
X_t, y_t = unison_shuffled_copies(X_t, y_t)
return X_t, y_t
def fun(*args, **kwargs):
return [load_events(*args, **kwargs)]