def multi_code_sklearn(arrs, base=None, method=None, init=None, c=None, params=None):
"""Multi-array Non-negative matrix factorization using scikits-learn.
- arrs(``path+``) input arrays (loci x scaled features) see: ``scale_features``.
- base(``str``) common basename for the output.
- c(``int``) number of expected histone codes (factorization rank).
- init(``str``) matrix factorization initialization method.
"""
kwargs = parse_params(params, {"max_iter":1000})
marks, xs = load_arrs(arrs)
hs = []
for x in xs:
nmf = decomposition.NMF(n_components=c, init=init, sparseness='components', **kwargs)
nmf.fit(x)
hs.append(nmf.components_)
H = np.vstack(hs)
X = np.vstack(xs)
W = np.zeros((X.shape[0], len(H)))
for j in range(0, X.shape[0]):
W[j, :], _ = nnls(H.T, X[j, :])
# write codes
ofnc = base + ("_%s-c#%s-i#%s-p#%s.epi" % ("pgnmf", c, init, (params or "")))
write_codes(ofnc, H, marks)
# write
ofna = base + ("_%s-c#%s-i#%s-p#%s.arr" % ("pgnmf", c, init, (params or "")))
write_values(ofna, W, len(arrs)*c)
return ofnc, ofna
def recode_sklearn(arr=None, epi=None, odn=path("."), base=None):
"""(internal) projects arr onto codes
- arr(``path``)
- epi(``path``)
"""
arr_marks, X = load_arr(arr)
epi_marks, H = load_epi(epi)
assert arr_marks == epi_marks
W = np.zeros((X.shape[0], len(H)))
for j in range(0, X.shape[0]):
W[j, :], _ = nnls(H.T, X[j, :])
base = base or arr.basename().splitext()[0] + "_" + epi.basename().splitext()[0]
ofn = odn / (base + ".arr")
# write
write_values(ofn, W, W.shape[1])
def recode_sklearn(arr=None, epi=None, odn=path("."), base=None):
"""(internal) projects arr onto codes
- arr(``path``)
- epi(``path``)
"""
arr_marks, X = load_arr(arr)
epi_marks, H = load_epi(epi)
assert arr_marks == epi_marks
W = np.zeros((X.shape[0], len(H)))
for j in range(0, X.shape[0]):
W[j, :], _ = nnls(H.T, X[j, :])
base = base or arr.basename().splitext()[0] + "_" + epi.basename(
).splitext()[0]
ofn = odn / (base + ".arr")
# write
write_values(ofn, W, W.shape[1])
def multi_code_sklearn(arrs,
base=None,
method=None,
init=None,
c=None,
params=None):
"""Multi-array Non-negative matrix factorization using scikits-learn.
- arrs(``path+``) input arrays (loci x scaled features) see: ``scale_features``.
- base(``str``) common basename for the output.
- c(``int``) number of expected histone codes (factorization rank).
- init(``str``) matrix factorization initialization method.
"""
kwargs = parse_params(params, {"max_iter": 1000})
marks, xs = load_arrs(arrs)
hs = []
for x in xs:
nmf = decomposition.NMF(n_components=c,
init=init,
sparseness='components',
**kwargs)
nmf.fit(x)
hs.append(nmf.components_)
H = np.vstack(hs)
X = np.vstack(xs)
W = np.zeros((X.shape[0], len(H)))
for j in range(0, X.shape[0]):
W[j, :], _ = nnls(H.T, X[j, :])
# write codes
ofnc = base + ("_%s-c#%s-i#%s-p#%s.epi" % ("pgnmf", c, init,
(params or "")))
write_codes(ofnc, H, marks)
# write
ofna = base + ("_%s-c#%s-i#%s-p#%s.arr" % ("pgnmf", c, init,
(params or "")))
write_values(ofna, W, len(arrs) * c)
return ofnc, ofna
