def KL_method(X, y, k_for_meaning = 5, epsilon = 0.01, begin = None, end = None, num = None, progress_bar = None, multiprocess=False):
if progress_bar is None:
progress_bar = list
else:
progress_bar=lambda x: tqdm(x, leave = False)
if end is None:
end = X.shape[0] - 1
if begin is None:
begin = 2*X.shape[1]
if num is None:
num = 5
subset_sizes = np.arange(begin, end, num, dtype=np.int64)
m_size = end
X, y = shuffle(X, y)
list_of_E, list_of_S = ut.calculater(X, y, subset_sizes, k_for_meaning, ut.klFunction, multiprocess, progress_bar)
for m, mean in zip(reversed(subset_sizes), reversed(list_of_E)):
if mean < epsilon:
m_size = m
return {'m*': m_size,
'E': np.array(list_of_E),
'S': np.array(list_of_S),
'm': np.array(subset_sizes),
}
def ALC(X, y, k_for_meaning = 100, l = 0.5, alpha = 0.05, begin = None, end = None, num = None, progress_bar = None, multiprocess = False):
if progress_bar is None:
progress_bar = list
else:
progress_bar=lambda x: tqdm(x, leave = False)
if end is None:
end = X.shape[0] - 1
if begin is None:
begin = 2*X.shape[1]
if num is None:
num = 5
subset_sizes = np.arange(begin, end, num, dtype=np.int64)
m_size = end
X, y = shuffle(X, y)
list_of_E, list_of_S = ut.calculater(X, y, subset_sizes, k_for_meaning, ut.aDistribution, multiprocess, progress_bar, alpha)
for m, mean, std in zip(reversed(subset_sizes), reversed(list_of_E), reversed(list_of_S)):
if mean < l:
m_size = m
return {'m*': m_size,
'E': np.array(list_of_E),
'S': np.array(list_of_S),
'm': np.array(subset_sizes),
}
def MAX_U(X, y, k_for_meaning = 100, c = 0.005, begin = None, end = None, num = None, progress_bar = None, multiprocess=False):
if progress_bar is None:
progress_bar = list
else:
progress_bar=lambda x: tqdm(x, leave = False)
if end is None:
end = X.shape[0] - 1
if begin is None:
begin = 2*X.shape[1]
if num is None:
num = max(5, int(X.shape[0]/20))
subset_sizes = np.arange(begin, end, num, dtype=np.int64)
m_size = end
X, y = shuffle(X, y)
list_of_E, list_of_S = ut.calculater(X, y, subset_sizes, k_for_meaning, ut.uFunction, multiprocess, progress_bar, c)
return {'m*': subset_sizes[np.argmax(np.array(list_of_E))],
'E': np.array(list_of_E),
'S': np.array(list_of_S),
'm': np.array(subset_sizes),
}