Python numpy 模块，reciprocal() 实例源码
我们从Python开源项目中，提取了以下50个代码示例，用于说明如何使用numpy.reciprocal()。
def guess(representation, sims, xi, a, a_, b):
sa = sims[xi[a]]
sa_ = sims[xi[a_]]
sb = sims[xi[b]]
add_sim = -sa+sa_+sb
if a in representation.wi:
add_sim[representation.wi[a]] = 0
if a_ in representation.wi:
add_sim[representation.wi[a_]] = 0
if b in representation.wi:
add_sim[representation.wi[b]] = 0
b_add = representation.iw[np.nanargmax(add_sim)]
mul_sim = sa_*sb*np.reciprocal(sa+0.01)
if a in representation.wi:
mul_sim[representation.wi[a]] = 0
if a_ in representation.wi:
mul_sim[representation.wi[a_]] = 0
if b in representation.wi:
mul_sim[representation.wi[b]] = 0
b_mul = representation.iw[np.nanargmax(mul_sim)]
return b_add, b_mul
def calc_pmi(counts, cds):
"""
Calculates e^PMI; PMI without the log().
"""
sum_w = np.array(counts.sum(axis=1))[:, 0]
sum_c = np.array(counts.sum(axis=0))[0, :]
if cds != 1:
sum_c = sum_c ** cds
sum_total = sum_c.sum()
sum_w = np.reciprocal(sum_w)
sum_c = np.reciprocal(sum_c)
pmi = csr_matrix(counts)
pmi = multiply_by_rows(pmi, sum_w)
pmi = multiply_by_columns(pmi, sum_c)
pmi = pmi * sum_total
return pmi
def _build_lowrank_op(U, s, V, Y):
"""
Private method that computes the lowrank operator from the singular
value decomposition of matrix X and the matrix Y.
.. math::
\\mathbf{\\tilde{A}} =
\\mathbf{U}^* \\mathbf{Y} \\mathbf{X}^\\dagger \\mathbf{U} =
\\mathbf{U}^* \\mathbf{Y} \\mathbf{V} \\mathbf{S}^{-1}
:param numpy.ndarray U: 2D matrix that contains the left-singular
vectors of X, stored by column.
:param numpy.ndarray s: 1D array that contains the singular values of X.
:param numpy.ndarray V: 2D matrix that contains the right-singular
vectors of X, stored by row.
:param numpy.ndarray Y: input matrix Y.
:return: the lowrank operator
:rtype: numpy.ndarray
"""
return U.T.conj().dot(Y).dot(V) * np.reciprocal(s)
def get_inverse_distance_matrix(self):
"""Calculates the inverse distance matrix A defined as:
A_ij = 1/|r_i - r_j|
For periodic systems the distance of an atom from itself is the
smallest displacement of an atom from one of it's periodic copies, and
the distance of two different atoms is the distance of two closest
copies.
Returns:
np.array: Symmetric 2D matrix containing the pairwise inverse
distances.
"""
if self._inverse_distance_matrix is None:
distance_matrix = self.get_distance_matrix()
with np.errstate(divide='ignore'):
inv_distance_matrix = np.reciprocal(distance_matrix)
self._inverse_distance_matrix = inv_distance_matrix
return self._inverse_distance_matrix
def pdf(cls, x, a, b):
"""Density function at `x`.
Parameters
----------
x : float or array-like
a : float or array-like
b : float or array-like
Returns
-------
np.array
"""
with np.errstate(divide='ignore'):
p = np.where((x < np.exp(a)) | (x > np.exp(b)), 0, np.reciprocal(x))
p /= (b - a) # normalize
return p
def XB(Y,index_PQ, index_P, n_PQ, n_P):
case_number, _ = np.shape(Y)
G, B = iterator.util.get_G_B(Y)
X = np.zeros((case_number, case_number))
B_p = np.zeros((n_P,n_P))
B_pp = np.zeros((n_PQ,n_PQ))
for i in xrange(case_number):
for j in xrange(case_number):
if LA.norm(Y[i][j]) > 1e-5 and i != j:
X[i][j] = np.reciprocal(np.imag(np.reciprocal(Y[i][j])))
for i in xrange(case_number):
for j in xrange(case_number):
if i != j:
X[i][i] -= X[i][j]
for i in xrange(0, n_P):
for j in xrange(0, n_P):
B_p[i][j] = X[index_P[i]][index_P[j]]
#---------------------------------------------------
for i in xrange(0, n_PQ):
for j in xrange(0, n_PQ):
B_pp[i][j] = B[index_PQ[i]][index_PQ[j]]
return B_p, B_pp
def comp_sum(vectors, reciprocal=False):
"""
:param vectors: vectors to be composed
:param reciprocal: if True, apply reciprocal weighting
:return: composed vector representation
"""
if not reciprocal:
composed_vector = np.sum(vectors, axis=0)
else:
weight_vector = np.reciprocal(np.arange(1., len(vectors) + 1))
weighted_vectors = []
for i, weight in enumerate(weight_vector):
weighted_vectors.append(vectors[i] * weight)
composed_vector = reduce(lambda x, y: x + y, weighted_vectors)
return composed_vector
def comp_mult(vectors, reciprocal=False):
"""
:param vectors: vectors to be composed
:param reciprocal: if True, apply reciprocal weighting
:return: composed vector representation
"""
if not reciprocal:
composed_vector = reduce(lambda x, y: x * y, vectors)
else:
weight_vector = np.reciprocal(np.arange(1., len(vectors) + 1))
weighted_vectors = []
for i, weight in enumerate(weight_vector):
weighted_vectors.append(vectors[i] * weight)
composed_vector = reduce(lambda x, y: x * y, weighted_vectors)
return composed_vector
def comp_max(vectors, reciprocal=False):
"""
:param vectors: vectors to be composed
:param reciprocal: if True, apply reciprocal weighting
:return: composed vector representation
"""
if not reciprocal:
composed_vector = np.amax(vectors, axis=0)
else:
weight_vector = np.reciprocal(np.arange(1., len(vectors) + 1))
weighted_vectors = []
for i, weight in enumerate(weight_vector):
weighted_vectors.append(vectors[i] * weight)
composed_vector = np.amax(weighted_vectors, axis=0)
return composed_vector
def calcuate_similarity(pivot_table, user_data, product_data, i, j, w_lambda):
if i==j:
return 0
normalize_freq = np.max(product_data['count'].values)
common = (pivot_table[i]*pivot_table[j]).nonzero()
rating_i = pivot_table[i][common[0]]
rating_j = pivot_table[j][common[0]]
rating_i = rating_i - user_data.iloc[i, 0]
rating_j = rating_j - user_data.iloc[j, 0]
variance = rating_i*rating_j
reputation = product_data.iloc[common[0], 0].as_matrix()/5
frequency = product_data.iloc[common[0], 2]/normalize_freq
val = np.sum(np.sqrt(w_lambda*np.square(np.reciprocal(reputation))+(1-w_lambda)*np.square(np.reciprocal(frequency)))*variance)
return val/ ( max(user_data.iloc[i, 1], 1)*max(user_data.iloc[j, 1], 1) )
#Read ratings data
def calcuate_similarity(pivot_table, user_data, product_data, i, j, w_lambda):
if i==j:
return 0
normalize_freq = np.max(product_data['count'].values)
common = (pivot_table[i]*pivot_table[j]).nonzero()
rating_i = pivot_table[i][common[0]]
rating_j = pivot_table[j][common[0]]
rating_i = rating_i - user_data.iloc[i, 0]
rating_j = rating_j - user_data.iloc[j, 0]
variance = rating_i*rating_j
reputation = product_data.iloc[common[0], 0].as_matrix()/5
frequency = product_data.iloc[common[0], 2]/normalize_freq
val = np.sum(np.sqrt(w_lambda*np.square(np.reciprocal(reputation))+(1-w_lambda)*np.square(np.reciprocal(frequency))))
return val/ ( max(user_data.iloc[i, 1], 1)*max(user_data.iloc[j, 1], 1) )
#create pandas dataframe
# df = pd.read_csv('dataset/ratings_Electronics_compressed.csv',
# header=None,
# names=['reviewerID', 'productID', 'overall', 'unixReviewTime'],
# sep=',',
# dtype={'reviewerID':int, 'productID':int, 'overall':int, 'unixReviewTime':int})
def calcuate_similarity(pivot_table, user_data, product_data, i, j, w_lambda):
if i==j:
return 0
normalize_freq = np.max(product_data['count'].values)
common = (pivot_table[i]*pivot_table[j]).nonzero()
rating_i = pivot_table[i][common[0]]
rating_j = pivot_table[j][common[0]]
rating_i = rating_i - user_data.iloc[i, 0]
rating_j = rating_j - user_data.iloc[j, 0]
variance = rating_i*rating_j
reputation = product_data.iloc[common[0], 0].as_matrix()/5
frequency = product_data.iloc[common[0], 2]/normalize_freq
val = np.sum(np.sqrt(w_lambda*np.square(np.reciprocal(reputation))+(1-w_lambda)*np.square(np.reciprocal(frequency)))*variance)
return val/ ( max(user_data.iloc[i, 1], 1)*max(user_data.iloc[j, 1], 1) )
#Read ratings data
def calcuate_similarity(pivot_table, user_data, product_data, i, j, w_lambda):
if i==j:
return 0
normalize_freq = np.max(product_data['count'].values)
common = (pivot_table[i]*pivot_table[j]).nonzero()
rating_i = pivot_table[i][common[0]]
rating_j = pivot_table[j][common[0]]
rating_i = rating_i - user_data.iloc[i, 0]
rating_j = rating_j - user_data.iloc[j, 0]
variance = rating_i*rating_j
reputation = product_data.iloc[common[0], 0].as_matrix()/5
frequency = product_data.iloc[common[0], 2]/normalize_freq
val = np.sum(np.sqrt(w_lambda*np.square(np.reciprocal(reputation))+(1-w_lambda)*np.square(np.reciprocal(frequency)))*variance)
return val/ ( max(user_data.iloc[i, 1], 1)*max(user_data.iloc[j, 1], 1) )
#Read ratings data
def calcuate_similarity(pivot_table, user_data, product_data, i, j, w_lambda):
if i==j:
return 0
normalize_freq = np.max(product_data['count'].values)
common = (pivot_table[i]*pivot_table[j]).nonzero()
rating_i = pivot_table[i][common[0]]
rating_j = pivot_table[j][common[0]]
rating_i = rating_i - user_data.iloc[i, 0]
rating_j = rating_j - user_data.iloc[j, 0]
variance = rating_i*rating_j
reputation = product_data.iloc[common[0], 0].as_matrix()/5
frequency = product_data.iloc[common[0], 2]/normalize_freq
val = np.sum(np.sqrt(w_lambda*np.square(np.reciprocal(reputation))+(1-w_lambda)*np.square(np.reciprocal(frequency)))*variance)
return val/ ( max(user_data.iloc[i, 1], 1)*max(user_data.iloc[j, 1], 1) )
#create pandas dataframe
# df = pd.read_csv('dataset/ratings_Electronics_compressed.csv',
# header=None,
# names=['reviewerID', 'productID', 'overall', 'unixReviewTime'],
# sep=',',
# dtype={'reviewerID':int, 'productID':int, 'overall':int, 'unixReviewTime':int})
def calcuate_similarity(pivot_table, user_data, product_data, i, j, w_lambda):
if i==j:
return 0
normalize_freq = np.max(product_data['count'].values)
common = (pivot_table[i]*pivot_table[j]).nonzero()
rating_i = pivot_table[i][common[0]]
rating_j = pivot_table[j][common[0]]
rating_i = rating_i - user_data.iloc[i, 0]
rating_j = rating_j - user_data.iloc[j, 0]
variance = rating_i*rating_j
reputation = product_data.iloc[common[0], 0].as_matrix()/5
frequency = product_data.iloc[common[0], 2]/normalize_freq
val = np.sum(np.sqrt(w_lambda*np.square(np.reciprocal(reputation))+(1-w_lambda)*np.square(np.reciprocal(frequency)))*variance)
return val/ ( max(user_data.iloc[i, 1], 1)*max(user_data.iloc[j, 1], 1) )
#Read ratings data
def _butterworth_filter(rows, cols, thresh, order):
# X and Y matrices with ranges normalised to +/- 0.5
array1 = np.ones(rows)
array2 = np.ones(cols)
array3 = np.arange(1,rows+1)
array4 = np.arange(1,cols+1)
x = np.outer(array1, array4)
y = np.outer(array3, array2)
x = x - float(cols/2) - 1
y = y - float(rows/2) - 1
x = x / cols
y = y / rows
radius = np.sqrt(np.square(x) + np.square(y))
matrix1 = radius/thresh
matrix2 = np.power(matrix1, 2*order)
f = np.reciprocal(1 + matrix2)
return f
def Mstep(dataSet,W):
(N, M) = np.shape(dataSet)
K = np.size(W,1)
# Each column of MU represents the mean of a cluster.
# So, for K clusters, there will be K columns of MU
# Each column,
# mu_k = (1/N_k)*sum_{1}^{N}{w_{ik}*x_i}
N_k = np.sum(W,0)
Alpha = N_k/np.sum(N_k)
Mu = dataSet.T.dot(W).dot(np.diag(np.reciprocal(N_k)))
# SIGMA is a 3-dimensional matrix of size MxMxK.
# It contains K covariances for each cluster
Sigma = np.zeros([M,M,K])
for k in range(K):
datMeanSub = dataSet.T - Mu[0:,k][None].T.dot(np.ones([1,N]))
Sigma[:,:,k] = (datMeanSub.dot(np.diag(W[0:,k])).dot(datMeanSub.T))/N_k[k]
return Alpha,Mu,Sigma
def Estep(dataSet,Alpha,Mu,Sigma):
# We will calculate the membership weight matrix W here. W is an
# NxK matrix where (i,j)th element represents the probability of
# ith data point to be a member of jth cluster given the parameters
# Alpha, Mu and Sigma
N = np.size(dataSet,0)
K = np.size(Alpha)
W = np.zeros([N,K])
for k in range(K):
for i in range(N):
W[i,k] = Alpha[k]*norm_pdf_multivariate(dataSet[i,:][None].T, \
Mu[:,k][None].T,Sigma[:,:,k])
# Normalize W row-wise because each row represents a pdf. In other words,
# probability of a point to be any one of the K clusters is equal to 1.
W = W*np.reciprocal(np.sum(W,1)[None].T)
return W
def guess(representation, sims, xi, a, a_, b):
sa = sims[xi[a]]
sa_ = sims[xi[a_]]
sb = sims[xi[b]]
add_sim = -sa+sa_+sb
if a in representation.wi:
add_sim[representation.wi[a]] = 0
if a_ in representation.wi:
add_sim[representation.wi[a_]] = 0
if b in representation.wi:
add_sim[representation.wi[b]] = 0
b_add = representation.iw[np.nanargmax(add_sim)]
mul_sim = sa_*sb*np.reciprocal(sa+0.01)
if a in representation.wi:
mul_sim[representation.wi[a]] = 0
if a_ in representation.wi:
mul_sim[representation.wi[a_]] = 0
if b in representation.wi:
mul_sim[representation.wi[b]] = 0
b_mul = representation.iw[np.nanargmax(mul_sim)]
return b_add, b_mul
def predict_proba(self, X):
"""Estimate probability.
Parameters
----------
X : array-like, shape (n_samples, n_features)
Input data.
Returns
-------
C : array, shape (n_samples, n_classes)
Estimated probabilities.
"""
prob = self.decision_function(X)
prob *= -1
np.exp(prob, prob)
prob += 1
np.reciprocal(prob, prob)
if len(self.classes_) == 2: # binary case
return np.column_stack([1 - prob, prob])
else:
# OvR normalization, like LibLinear's predict_probability
prob /= prob.sum(axis=1).reshape((prob.shape[0], -1))
return prob
def _predict_proba_lr(self, X):
"""Probability estimation for OvR logistic regression.
Positive class probabilities are computed as
1. / (1. + np.exp(-self.decision_function(X)));
multiclass is handled by normalizing that over all classes.
"""
prob = self.decision_function(X)
prob *= -1
np.exp(prob, prob)
prob += 1
np.reciprocal(prob, prob)
if prob.ndim == 1:
return np.vstack([1 - prob, prob]).T
else:
# OvR normalization, like LibLinear's predict_probability
prob /= prob.sum(axis=1).reshape((prob.shape[0], -1))
return prob
def __resampling(self, px, pw):
'''?????????
?????????(ESS)?????????????????????????
???
px??????????
pw??????????
????
px???????????????
pw???????????????
'''
ess = float(np.reciprocal(pw @ pw.T))
if ess < self.__ESS_TH:
pw_cum = np.cumsum(pw)
base_id = np.arange(0.0, 1.0, self.__NP_RECIP)
ofs = np.random.rand() * self.__NP_RECIP # ?????????
resample_id = base_id + ofs
px_temp = np.copy(px)
idx = 0
for i in range(self.__NP):
while resample_id[i] > pw_cum[idx]:
idx += 1
px[:, i] = px_temp[:, idx]
pw = np.copy(self.__pw_ini) # ??????
return px, pw
def compute_svd_normalization(samples, Ndiscard, max_rescale):
# S is list of singular values in descending order
# Each row of V is a list of the weights of the features in a given principal component
centered_samples = samples - samples.mean(axis=0) # subtract columnwise means
U, S, V = np.linalg.svd(centered_samples, full_matrices=False)
Nsamp = samples.shape[0]
component_stddevs = S / math.sqrt(Nsamp)
print "singular values ="
print S
print "component standard deviations ="
print component_stddevs
print "V matrix ="
print V
Nfeat = samples.shape[1]
rescaling_factors = np.minimum(np.reciprocal(component_stddevs[:Nfeat-Ndiscard]), max_rescale)
whitening_matrix = np.dot(V[:Nfeat-Ndiscard].T, np.diag(rescaling_factors))
print "Ndiscard =", Ndiscard
print "max_rescale =", max_rescale
print "rescaling_factors ="
print rescaling_factors
print "whitening_matrix ="
print repr(whitening_matrix)
def normalize(self):
m2 = self.m.copy()
m2.data **= 2
norm = np.reciprocal(np.sqrt(np.array(m2.sum(axis=1))[:, 0]))
normalizer = dok_matrix((len(norm), len(norm)))
normalizer.setdiag(norm)
self.m = normalizer.tocsr().dot(self.m)
def test_blocked(self):
# test alignments offsets for simd instructions
# alignments for vz + 2 * (vs - 1) + 1
for dt, sz in [(np.float32, 11), (np.float64, 7)]:
for out, inp1, inp2, msg in _gen_alignment_data(dtype=dt,
type='binary',
max_size=sz):
exp1 = np.ones_like(inp1)
inp1[...] = np.ones_like(inp1)
inp2[...] = np.zeros_like(inp2)
assert_almost_equal(np.add(inp1, inp2), exp1, err_msg=msg)
assert_almost_equal(np.add(inp1, 1), exp1 + 1, err_msg=msg)
assert_almost_equal(np.add(1, inp2), exp1, err_msg=msg)
np.add(inp1, inp2, out=out)
assert_almost_equal(out, exp1, err_msg=msg)
inp2[...] += np.arange(inp2.size, dtype=dt) + 1
assert_almost_equal(np.square(inp2),
np.multiply(inp2, inp2), err_msg=msg)
assert_almost_equal(np.reciprocal(inp2),
np.divide(1, inp2), err_msg=msg)
inp1[...] = np.ones_like(inp1)
inp2[...] = np.zeros_like(inp2)
np.add(inp1, 1, out=out)
assert_almost_equal(out, exp1 + 1, err_msg=msg)
np.add(1, inp2, out=out)
assert_almost_equal(out, exp1, err_msg=msg)
def _eig_from_lowrank_op(Atilde, Y, U, s, V, exact):
"""
Private method that computes eigenvalues and eigenvectors of the
high-dimensional operator from the low-dimensional operator and the
input matrix.
:param numpy.ndarray Atilde: the lowrank operator.
:param numpy.ndarray Y: input matrix Y.
:param numpy.ndarray U: 2D matrix that contains the left-singular
vectors of X, stored by column.
:param numpy.ndarray s: 1D array that contains the singular values of X.
:param numpy.ndarray V: 2D matrix that contains the right-singular
vectors of X, stored by row.
:param bool exact: if True, the exact modes are computed; otherwise,
the projected ones are computed.
:return: eigenvalues, eigenvectors
:rtype: numpy.ndarray, numpy.ndarray
"""
lowrank_eigenvalues, lowrank_eigenvectors = np.linalg.eig(Atilde)
# Compute the eigenvectors of the high-dimensional operator
if exact:
eigenvectors = (
(Y.dot(V) * np.reciprocal(s)).dot(lowrank_eigenvectors)
)
else:
eigenvectors = U.dot(lowrank_eigenvectors)
# The eigenvalues are the same
eigenvalues = lowrank_eigenvalues
return eigenvalues, eigenvectors
def sine_matrix(self, system):
"""Creates the Sine matrix for the given system.
"""
# Cell and inverse cell
B = system.get_cell()
B_inv = system.get_cell_inverse()
# Difference vectors in tensor 3D-tensor-form
diff_tensor = system.get_displacement_tensor()
# Calculate phi
arg_to_sin = np.pi * np.dot(diff_tensor, B_inv)
phi = np.linalg.norm(np.dot(np.sin(arg_to_sin)**2, B), axis=2)
with np.errstate(divide='ignore'):
phi = np.reciprocal(phi)
# Calculate Z_i*Z_j
q = system.get_initial_charges()
qiqj = q[None, :]*q[:, None]
np.fill_diagonal(phi, 0)
# Multiply by charges
smat = qiqj*phi
# Set diagonal
np.fill_diagonal(smat, 0.5 * q ** 2.4)
return smat
def test_reciprocal(input_tensor):
"""TODO."""
p_u = input_tensor
u = rng.uniform(.1, 5.0, p_u.axes)
rec_u_np = np.reciprocal(u)
rec_u = ng.reciprocal(p_u)
with ExecutorFactory() as ex:
rec_u_graph = ex.executor(rec_u, p_u)(u)
ng.testing.assert_allclose(rec_u_np, rec_u_graph)
def test_reciprocal_derivative(input_tensor):
"""TODO."""
p_u = input_tensor
delta = .001
u = rng.uniform(.1, 5.0, p_u.axes)
rec_u = ng.reciprocal(p_u)
check_derivative(rec_u, p_u, delta, u, atol=1e-2, rtol=1e-2)
def test_variance_sqrt_inverse(input_tensor):
inputs = input_tensor
targets = ng.placeholder(inputs.axes)
epsilon = 1e-3
inp_stat = ng.reciprocal(
ng.sqrt(
ng.variance(inputs, reduction_axes=inputs.axes.batch_axes()) + epsilon
)
)
err = ng.sum(inp_stat - targets, out_axes=())
d_inputs = ng.deriv(err, inputs)
with executor([err, d_inputs], inputs, targets) as comp_func:
input_value = rng.uniform(-1, 1, inputs.axes)
target_value = rng.uniform(-1, 1, targets.axes)
ng_f_res, ng_b_res = comp_func(input_value, target_value)
npv = np.var(input_value, axis=1, keepdims=True) + epsilon
np_f_res = 1.0 / np.sqrt(npv)
npv_delta = 2 * (input_value - np.mean(input_value, axis=1, keepdims=True))
np_b_res = - 0.5 * np_f_res / npv * npv_delta
np_f_res = np.sum(np_f_res - target_value)
ng.testing.assert_allclose(np_f_res, ng_f_res, atol=1e-4, rtol=1e-4)
ng.testing.assert_allclose(np_b_res, ng_b_res, atol=1e-4, rtol=1e-4)
def test_reciprocal(input_data):
expected_output = np.reciprocal(input_data)
node = onnx.helper.make_node('Reciprocal', inputs=['x'], outputs=['y'])
ng_results = convert_and_calculate(node, [input_data], [expected_output])
assert np.allclose(ng_results, [expected_output])
def testCplxReciprocalGPU(self):
shapes = [(5,4,3), (5,4), (5,), (1,)]
for sh in shapes:
x = ((np.random.randn(*sh) +
1j*np.random.randn(*sh)).astype(np.complex64))
self._compareGpu(x, np.reciprocal, tf.reciprocal)
def testCplxReciprocalGradGPU(self):
shapes = [(5,4,3), (5,4), (5,), (1,)]
for sh in shapes:
x = ((np.random.randn(*sh) +
1j*np.random.randn(*sh)).astype(np.complex64))
self._compareGpuGrad(x, np.reciprocal, tf.reciprocal)
def test_blocked(self):
# test alignments offsets for simd instructions
# alignments for vz + 2 * (vs - 1) + 1
for dt, sz in [(np.float32, 11), (np.float64, 7)]:
for out, inp1, inp2, msg in _gen_alignment_data(dtype=dt,
type='binary',
max_size=sz):
exp1 = np.ones_like(inp1)
inp1[...] = np.ones_like(inp1)
inp2[...] = np.zeros_like(inp2)
assert_almost_equal(np.add(inp1, inp2), exp1, err_msg=msg)
assert_almost_equal(np.add(inp1, 1), exp1 + 1, err_msg=msg)
assert_almost_equal(np.add(1, inp2), exp1, err_msg=msg)
np.add(inp1, inp2, out=out)
assert_almost_equal(out, exp1, err_msg=msg)
inp2[...] += np.arange(inp2.size, dtype=dt) + 1
assert_almost_equal(np.square(inp2),
np.multiply(inp2, inp2), err_msg=msg)
assert_almost_equal(np.reciprocal(inp2),
np.divide(1, inp2), err_msg=msg)
inp1[...] = np.ones_like(inp1)
inp2[...] = np.zeros_like(inp2)
np.add(inp1, 1, out=out)
assert_almost_equal(out, exp1 + 1, err_msg=msg)
np.add(1, inp2, out=out)
assert_almost_equal(out, exp1, err_msg=msg)
def BX(Y,index_PQ, index_P, n_PQ, n_P):
case_number, _ = np.shape(Y)
Y_p = Y.copy()
X_p = np.zeros((case_number, case_number))
B_p = np.zeros((n_P,n_P))
B_pp = np.zeros((n_PQ,n_PQ))
#-------------------------------------------------
for i in xrange(case_number):
Y_p[i][i] = complex(0,0)
for j in xrange(case_number):
if i != j:
Y_p[i][i] -= Y_p[i][j]
B = np.imag(Y_p)
for i in xrange(n_P):
for j in xrange(n_P):
B_p[i][j] = B[index_P[i]][index_P[j]]
#-------------------------------------------------
g_b_round = np.zeros(case_number)
for i in xrange(case_number):
a = np.sum(Y[i])
if LA.norm(a) > 1e-5:
g_b_round[i] = np.reciprocal(np.imag(np.reciprocal(a)))
for i in xrange(case_number):
for j in xrange(case_number):
if LA.norm(Y[i][j]) > 1e-5 and i!=j:
X_p[i][j] = np.reciprocal(np.imag(np.reciprocal(Y[i][j])))
for i in xrange(case_number):
X_p[i][i] = g_b_round[i]
for j in xrange(case_number):
if i != j:
X_p[i][i] -= X_p[i][j]
for i in xrange(0, n_PQ):
for j in xrange(0, n_PQ):
B_pp[i][j] = X_p[index_PQ[i]][index_PQ[j]]
return B_p, - B_pp
# Stott ================================================================================================================
# Stott Original--------------------------------------------------------------------------------------------------------
def XB(Y,index_PQ, index_P, n_PQ, n_P):
case_number, _ = np.shape(Y)
_, B = util.get_G_B(Y)
X = np.zeros((case_number, case_number))
B_p = np.zeros((n_P,n_P))
B_pp = np.zeros((n_PQ,n_PQ))
for i in xrange(case_number):
for j in xrange(case_number):
if LA.norm(Y[i][j]) > 1e-5 and i != j:
X[i][j] = np.reciprocal(np.imag(np.reciprocal(Y[i][j])))
for i in xrange(case_number):
for j in xrange(case_number):
if i != j:
X[i][i] -= X[i][j]
for i in xrange(0, n_P):
for j in xrange(0, n_P):
B_p[i][j] = X[index_P[i]][index_P[j]]
#---------------------------------------------------
for i in xrange(0, n_PQ):
for j in xrange(0, n_PQ):
B_pp[i][j] = B[index_PQ[i]][index_PQ[j]]
return - B_p, B_pp
# Stott count r in B'---------------------------------------------------------------------------------------------------
def XB_ground(Y,index_PQ, index_P, n_PQ, n_P):
case_number, _ = np.shape(Y)
_, B = util.get_G_B(Y)
X_p = np.zeros((case_number, case_number))
B_p = np.zeros((n_P, n_P))
B_pp = np.zeros((n_PQ, n_PQ))
g_b_round = np.zeros(case_number)
for i in xrange(case_number):
a = np.sum(Y[i])
if LA.norm(a) > 1e-5:
g_b_round[i] = np.reciprocal(np.imag(np.reciprocal(a)))
for i in xrange(case_number):
for j in xrange(case_number):
if LA.norm(Y[i][j]) > 1e-5 and i!=j:
X_p[i][j] = np.reciprocal(np.imag(np.reciprocal(Y[i][j])))
for i in xrange(case_number):
X_p[i][i] = g_b_round[i]
for j in xrange(case_number):
if i != j:
X_p[i][i] -= X_p[i][j]
for i in xrange(0, n_P):
for j in xrange(0, n_P):
B_p[i][j] = X_p[index_P[i]][index_P[j]]
# ---------------------------------------------------
for i in xrange(0, n_PQ):
for j in xrange(0, n_PQ):
B_pp[i][j] = B[index_PQ[i]][index_PQ[j]]
return - B_p, B_pp
# Stott ================================================================================================================
# XX--------------------------------------------------------------------------------------------------------------------
def XX(Y,index_PQ, index_P, n_PQ, n_P):
case_number, _ = np.shape(Y)
X_p = np.zeros((case_number, case_number))
B_p = np.zeros((n_P,n_P))
B_pp = np.zeros((n_PQ,n_PQ))
for i in xrange(case_number):
for j in xrange(case_number):
if LA.norm(Y[i][j]) > 1e-5 and i != j:
X_p[i][j] = np.reciprocal(np.imag(np.reciprocal(Y[i][j])))
for i in xrange(case_number):
for j in xrange(case_number):
if i != j:
X_p[i][i] -= X_p[i][j]
for i in xrange(0, n_P):
for j in xrange(0, n_P):
B_p[i][j] = X_p[index_P[i]][index_P[j]]
#-------------------------------------------------
g_b_round = np.zeros(case_number)
for i in xrange(case_number):
a = np.sum(Y[i])
if LA.norm(a) > 1e-5:
g_b_round[i] = np.reciprocal(np.imag(np.reciprocal(a)))
for i in xrange(case_number):
for j in xrange(case_number):
if LA.norm(Y[i][j]) > 1e-5 and i!=j:
X_p[i][j] = np.reciprocal(np.imag(np.reciprocal(Y[i][j])))
for i in xrange(case_number):
X_p[i][i] = g_b_round[i]
for j in xrange(case_number):
if i != j:
X_p[i][i] -= X_p[i][j]
for i in xrange(0, n_PQ):
for j in xrange(0, n_PQ):
B_pp[i][j] = X_p[index_PQ[i]][index_PQ[j]]
return - B_p, - B_pp
def BX(Y,index_PQ, index_P, n_PQ, n_P):
case_number, _ = np.shape(Y)
Y_p = Y.copy()
# print Y
X_p = np.zeros((case_number, case_number))
B_p = np.zeros((n_P,n_P))
B_pp = np.zeros((n_PQ,n_PQ))
#-------------------------------------------------
for i in xrange(case_number):
Y_p[i][i] = complex(0,0)
for j in xrange(case_number):
if i != j:
Y_p[i][i] -= Y_p[i][j]
B = np.imag(Y_p)
for i in xrange(n_P):
for j in xrange(n_P):
B_p[i][j] = B[index_P[i]][index_P[j]]
#-------------------------------------------------
g_b_round = np.zeros(case_number)
for i in xrange(case_number):
a = np.sum(Y[i])
if LA.norm(a) > 1e-5:
g_b_round[i] = np.reciprocal(np.imag(np.reciprocal(a)))
for i in xrange(case_number):
for j in xrange(case_number):
if LA.norm(Y[i][j]) > 1e-5 and i!=j:
X_p[i][j] = np.reciprocal(np.imag(np.reciprocal(Y[i][j])))
for i in xrange(case_number):
X_p[i][i] = g_b_round[i]
for j in xrange(case_number):
if i != j:
X_p[i][i] -= X_p[i][j]
for i in xrange(0, n_PQ):
for j in xrange(0, n_PQ):
B_pp[i][j] = X_p[index_PQ[i]][index_PQ[j]]
return B_p, B_pp
def __init__(self, parameters, language):
assert language in ["en", "nl"]
self.language = language
# load frequency list
pathtofrequencies = 'frequencies_' + language + '.json'
# load trained fasttext model
pathtomodel = 'embeddings_' + language + '.bin'
# give path to fasttext vectors
pathtovectors = 'embeddings_' + language + '.vec'
# PHASE 1
self.comp_function = parameters['comp_function'] # item from ["sum", "mult", "max"]
self.include_misspelling = parameters['include_misspelling'] # boolean
self.include_oov_candidates = parameters['include_oov_candidates'] # boolean
self.pathtovectors = pathtovectors # path to fasttext vectors
self.model = fasttext.load_model(pathtomodel) # path to fasttext model
# PHASE 2
self.window_size = parameters['window_size'] # number in range(0,11)
self.reciprocal = parameters['reciprocal'] # boolean
self.remove_stopwords = parameters['remove_stopwords'] # boolean
self.stopwords = frozenset(json.load(open('stopwords_' + str(self.language) + '.json', 'r')))
# PHASE 3
self.edit_distance = parameters['edit_distance'] # item from [1, 2, 3, 4]
# PHASE 4
self.oov_penalty = parameters['oov_penalty'] # oov penalty tuned with self.tune_oov()
# OUTPUT
self.ranking_method = parameters['ranking_method'] # item from ["context", "noisy_channel", "frequency",
# "ensemble"]
self.frequency_dict = json.load(open(pathtofrequencies, 'r')) # path to frequency list
self.k = parameters['k-best'] # positive natural number
def test_blocked(self):
# test alignments offsets for simd instructions
# alignments for vz + 2 * (vs - 1) + 1
for dt, sz in [(np.float32, 11), (np.float64, 7)]:
for out, inp1, inp2, msg in _gen_alignment_data(dtype=dt,
type='binary',
max_size=sz):
exp1 = np.ones_like(inp1)
inp1[...] = np.ones_like(inp1)
inp2[...] = np.zeros_like(inp2)
assert_almost_equal(np.add(inp1, inp2), exp1, err_msg=msg)
assert_almost_equal(np.add(inp1, 1), exp1 + 1, err_msg=msg)
assert_almost_equal(np.add(1, inp2), exp1, err_msg=msg)
np.add(inp1, inp2, out=out)
assert_almost_equal(out, exp1, err_msg=msg)
inp2[...] += np.arange(inp2.size, dtype=dt) + 1
assert_almost_equal(np.square(inp2),
np.multiply(inp2, inp2), err_msg=msg)
assert_almost_equal(np.reciprocal(inp2),
np.divide(1, inp2), err_msg=msg)
inp1[...] = np.ones_like(inp1)
inp2[...] = np.zeros_like(inp2)
np.add(inp1, 1, out=out)
assert_almost_equal(out, exp1 + 1, err_msg=msg)
np.add(1, inp2, out=out)
assert_almost_equal(out, exp1, err_msg=msg)
def KernelRadial(self, r):
return np.reciprocal(np.sqrt(1+self.gamma*r))
def predict_probas(X, w, intercept, multinomial=True):
"""
Predict probabilities for each class using either a multinomial or a
one-vs-rest approach
"""
#print X.shape
#print w.shape
#print intercept.shape
p = safe_sparse_dot(X, w.T, dense_output=True) + intercept
if multinomial:
return softmax(p, copy=False)
else:
p = p.ravel() if p.shape[1] == 1 else p
p *= -1
np.exp(p, p)
p += 1
np.reciprocal(p, p)
if p.ndim == 1:
return np.vstack([1 - p, p]).T
else:
# OvR normalization, like LibLinear's predict_probability
p /= p.sum(axis=1).reshape((p.shape[0], -1))
return p
def test_blocked(self):
# test alignments offsets for simd instructions
# alignments for vz + 2 * (vs - 1) + 1
for dt, sz in [(np.float32, 11), (np.float64, 7)]:
for out, inp1, inp2, msg in _gen_alignment_data(dtype=dt,
type='binary',
max_size=sz):
exp1 = np.ones_like(inp1)
inp1[...] = np.ones_like(inp1)
inp2[...] = np.zeros_like(inp2)
assert_almost_equal(np.add(inp1, inp2), exp1, err_msg=msg)
assert_almost_equal(np.add(inp1, 1), exp1 + 1, err_msg=msg)
assert_almost_equal(np.add(1, inp2), exp1, err_msg=msg)
np.add(inp1, inp2, out=out)
assert_almost_equal(out, exp1, err_msg=msg)
inp2[...] += np.arange(inp2.size, dtype=dt) + 1
assert_almost_equal(np.square(inp2),
np.multiply(inp2, inp2), err_msg=msg)
assert_almost_equal(np.reciprocal(inp2),
np.divide(1, inp2), err_msg=msg)
inp1[...] = np.ones_like(inp1)
inp2[...] = np.zeros_like(inp2)
np.add(inp1, 1, out=out)
assert_almost_equal(out, exp1 + 1, err_msg=msg)
np.add(1, inp2, out=out)
assert_almost_equal(out, exp1, err_msg=msg)
def test_blocked(self):
# test alignments offsets for simd instructions
# alignments for vz + 2 * (vs - 1) + 1
for dt, sz in [(np.float32, 11), (np.float64, 7)]:
for out, inp1, inp2, msg in _gen_alignment_data(dtype=dt,
type='binary',
max_size=sz):
exp1 = np.ones_like(inp1)
inp1[...] = np.ones_like(inp1)
inp2[...] = np.zeros_like(inp2)
assert_almost_equal(np.add(inp1, inp2), exp1, err_msg=msg)
assert_almost_equal(np.add(inp1, 1), exp1 + 1, err_msg=msg)
assert_almost_equal(np.add(1, inp2), exp1, err_msg=msg)
np.add(inp1, inp2, out=out)
assert_almost_equal(out, exp1, err_msg=msg)
inp2[...] += np.arange(inp2.size, dtype=dt) + 1
assert_almost_equal(np.square(inp2),
np.multiply(inp2, inp2), err_msg=msg)
assert_almost_equal(np.reciprocal(inp2),
np.divide(1, inp2), err_msg=msg)
inp1[...] = np.ones_like(inp1)
inp2[...] = np.zeros_like(inp2)
np.add(inp1, 1, out=out)
assert_almost_equal(out, exp1 + 1, err_msg=msg)
np.add(1, inp2, out=out)
assert_almost_equal(out, exp1, err_msg=msg)
def sigmoid(x, out):
if out is not x:
out[:] = x
np.negative(out, out)
np.exp(out, out)
out += 1
np.reciprocal(out, out)
return out
def calc_pmi(counts, cds):
sum_w = np.array(counts.sum(axis=1))[:, 0]
sum_c = np.array(counts.sum(axis=0))[0, :]
if cds != 1:
sum_c = sum_c ** cds
sum_total = sum_c.sum()
sum_w = np.reciprocal(sum_w)
sum_c = np.reciprocal(sum_c)
pmi = csr_matrix(counts)
pmi = multiply_by_rows(pmi, sum_w)
pmi = multiply_by_columns(pmi, sum_c)
pmi = pmi * sum_total
return pmi
def normalize(self):
m2 = self.m.copy()
m2.data **= 2
norm = np.reciprocal(np.sqrt(np.array(m2.sum(axis=1))[:, 0]))
normalizer = dok_matrix((len(norm), len(norm)))
normalizer.setdiag(norm)
self.m = normalizer.tocsr().dot(self.m)