def summer(idx_vals, idx_cols):
idxs = [i[0] for i in idx_vals]
if len(idxs) == 1:
return np.Vector(*idx_cols[idxs[0]])
else:
mutate_idxs = list(idx_vals)
tail = mutate_idxs[-1][0]
head = tuple(mutate_idxs[:-1])
return np.Vector(*idx_cols[tail]) + summer(head, idx_cols)
def fit(self, X, y):
cols = len(X[0])
rows = len(X)
beta = np.Vector(*cols * [0])
iteration = 0
XT = t(X)
while iteration < self.max_iter:
old_beta = np.Vector(*beta)
for j in range(len(beta)):
beta = self.step(X, XT, beta, y, j)
iteration += 1
if converged(beta, old_beta):
break
self.coef_ = beta
return self
def b1(size):
def vec(size):
return [urandom.getrandbits(8) / 100 for i in range(size)]
mat = (vec(size) for i in range(size))
v = np.Vector(*vec(size))
res = v.gen_matrix_mult(mat)
return
def step(self, X, XT, beta, y, j):
tmp_beta = np.change_idx(beta, j, 0.0)
r_j = y - self.cached_mult(tmp_beta, XT, recursive=True)
arg1 = r_j.inner(XT[j])
arg2 = self.alpha * len(X)
new_beta_j = self._soft_thresholding_operator(arg1, arg2) / sum(
np.Vector(*XT[j]).__pow__(2))
return np.change_idx(beta, j, new_beta_j)
def cached_mult(self, tmp_beta, XT, recursive=False):
active_coefs = get_active(tmp_beta)
if active_coefs:
if recursive:
all_cols = dict(
((zipped_idx_val[0], row_multer(zipped_idx_val, XT))
for zipped_idx_val in active_coefs))
idxs = tuple(active_coefs)
summed_cols = group_summer(idxs, all_cols)
else:
all_cols = (row_multer(zipped_idx_val, XT)
for zipped_idx_val in active_coefs)
summed_cols = np.Vector(*(sum(i) for i in zip(*all_cols))) #
return summed_cols
else:
dummy_lst = len(XT[0]) * [0]
summed_cols = np.Vector(*dummy_lst)
return summed_cols
def isft(spectrum):
inverse_W = ft.inv_dft_mat(len(spectrum))
v = np.Vector(*spectrum)
return spectrum.matrix_mult(inverse_W)
def mem_friendly_sft(signal):
N = len(signal)
W = ft.bp_fourier_generator(N)
v = np.Vector(*signal)
return v.gen_matrix_mult(W)
def sft(signal):
N = len(signal)
W = ft.bp_fourier(N)
v = np.Vector(*signal)
return v.matrix_mult(W)
def inverse_ft(spectrum):
N = len(spectrum) + 1
W = ft.bp_fourier(N)
vec = np.Vector(*spectrum)
return vec.matrix_mult(W)
def sparseft(signal, alpha=1e-4):
N = len(signal)
W = ft.bp_fourier(N)
v = np.Vector(*signal).zero_mean_normalize()
spectrum = sk_bp(W, v, alpha)
return spectrum
def atom_norm(atom):
return list(np.Vector(*atom).zero_mean_normalize())
def intra_summer(idx_vals, idx_cols):
all_cols = (idx_cols[v[0]] for v in idx_vals)
summed_cols = np.Vector(*(sum(i) for i in zip(*all_cols))) #
return summed_cols
