def _parse_dicom_and_contour_files(filenames):
"""Convert two filenames to valid image data
:param filenames: a 3-tuple record of dicom_filename, icontour_filename,
and ocontour_filename:
dicom_filename is the path string to the DICOM file
icontour_filename is the path string to the i-contour file
ocontour_filename is the path string to the o-contour file
:return: 5-tuple (RecordData) containing the DICOM image data and contour
mask data
"""
dicom_filename, icontour_filename, ocontour_filename = filenames
dicom_data, icontour_data, ocontour_data = None, None, None
icontour_path, ocontour_path = [], []
if dicom_filename:
dicom_data = parsing.parse_dicom_file(dicom_filename)
if dicom_data is not None:
dicom_data = dicom_data['pixel_data']
if icontour_filename:
icontour_path = parsing.parse_contour_file(icontour_filename)
if dicom_data is not None:
height, width = dicom_data.shape
else:
max_x, max_y = 0, 0
for x, y in icontour_path:
max_x = max(max_x, x)
max_y = max(max_y, y)
height = round(max_x + 1)
width = round(max_y + 1)
icontour_data = parsing.poly_to_mask(icontour_path, width, height)
if ocontour_filename:
ocontour_path = parsing.parse_contour_file(ocontour_filename)
if dicom_data is not None:
height, width = dicom_data.shape
else:
max_x, max_y = 0, 0
for x, y in ocontour_path:
max_x = max(max_x, x)
max_y = max(max_y, y)
height = round(max_x + 1)
width = round(max_y + 1)
ocontour_data = parsing.poly_to_mask(ocontour_path, width, height)
# TODO: fix the case in which all of them are None
if dicom_data is not None:
if icontour_data is None:
icontour_data = np.zeros(dicom_data.shape, dtype=np.bool_)
if ocontour_data is None:
ocontour_data = np.zeros(dicom_data.shape, dtype=np.bool_)
else:
if icontour_data is not None:
dicom_data = np.zeors(icontour_data.shape, dtype=np.int16)
elif ocontour_data is not None:
dicom_data = np.zeors(ocontour_data.shape, dtype=np.int16)
return RecordData(dicom_data, icontour_data, ocontour_data, icontour_path,
ocontour_path)
def test_create():
print("***常用创建****************************************")
arr1 = np.arange(5, dtype=float)
print('arange create:', arr1)
arr1 = np.linspace(10, 20, 5, endpoint=False)
print('arange linspace:', arr1)
print('zeros1:', np.zeors((5)))
print('zeros2:', np.zeors((5, 4)))
print('ones1:', np.ones((5, 4, 3)))
print('empty1:', np.empty((5, 4)))
print('full1:', np.full((5, 4), 3))
print('eye1:', np.eye(3, 4))
print('linspace1:', np.linspace(0, 10, num=4))
def costFunc(nn_params, input_layer_size, hidden_layer_size, num_labels, X, y, nn_lambda):
Theta1 = np.reshape(nn_params[0:(hidden_layer_size*(input_layer_size + 1))], (hidden_layer_size, input_layer_size + 1))
Theta2 = np.reshape(nn_params[hidden_layer_size*(input_layer_size + 1):np.size(nn_params)])
m = np.size(X,0)
J = 0
Theta1_grad = np.zeros((np.size(Theta1, 0), np.size(Theta1, 1)))
Theta2_grad = np.zeros((np.size(Theta2, 0), np.size(Theta2, 1)))
# feed forward
a1_all = np.concatenate((np.ones((m, 1)), X), axis=1)
z2_all = np.dot(Theta1, a1_all.transpose()).transpose()
a2_all = np.concatenate((np.ones((m, 1)), sigmoid(z2_all)), axis=1)
z3_all = np.dot(Theta2, a2_all.transpose()).transpose()
a3_all = sigmoid(z3_all)
all_labels = np.eye(num_labels)
# calculate 1st term corresponding to non regularized cost
s = 0
for cnt in range(0, m):
for cnt1 in range(1, num_labels + 1):
temp = -all_labels[y(cnt), cnt1]*np.log(a3_all[cnt, cnt1]) - (1 - all_labels[y(cnt), cnt1])*np.log(1 - a3_all[cnt, cnt1])
s = s + temp
J_nr = 1/m*s
# calculate 2nd tern corresponding to regularized part
# remember that we do not regularize bias intercepter
Theta1[:, 1] = 0
Theta2[:, 1] = 0
J_rg = nn_lambda / 2 / m * (np.sum(Theta1 ** 2) + np.sum(Theta2 ** 2))
J = J_nr + J_rg
# calculate gradient
delta3 = np.zeors((m, num_labels))
delta2 = np.zeors((m, hidden_layer_size))
D1 = np.zeros((np.size(Theta1, 0), np.size(Theta1, 1)))
D2 = np.zeros((np.size(Theta2, 0), np.size(Theta2, 1)))
for cnt in range(0, m):
delta3[cnt, :] = a3_all[cnt, :] - all_labels[y(cnt), :]
delta2[cnt, :] = np.dot(delta3[cnt, :], Theta2[:, 1:(hidden_layer_size + 1)]) * sigmoid_gradient(z2_all[cnt, :])
D1 = D1 + np.dot(delta2[cnt, :].transpose(), a1_all[cnt, :])
D2 = D2 + np.dot(delta3[cnt, :].transpose(), a2_all[cnt, :])
Theta1_grad = 1/m*D1
Theta2_grad = 1/m*D2
Theta1_grad[:, 1:(input_layer_size + 1)] = Theta1_grad[:, 1:(input_layer_size + 1)] + nn_lambda/m*Theta1[:, 1:(input_layer_size + 1)]
Theta2_grad[:, 1:(input_layer_size + 1)] = Theta2_grad[:, 1:(input_layer_size + 1)] + nn_lambda/m*Theta2[:, 1:(input_layer_size + 1)]
grad = np.concatenate((np.ravel(Theta1_grad), np.ravel(Theta2_grad)))
return J, grad
def projection(img):
h, w = img.shape[:2]
proj = np.zeors(h)
for i in range(w):
for j in range(h):
if img[j, i] > 200:
proj[j] += 1
def __init__(self, D):
self.running_mean = tf.Variable(np.zeros(D, dtype=np.float32),
trainable=True)
self.running_var = tf.Variable(np.zeros(D, dtype=np.float32),
trainable=True)
self.beta = tf.Variable(np.zeors(D, dtype=np.float32))
self.gamma = tf.Variable(np.zeros(D, dtype=np.float32))
def pick_strategy(self):
'''
TODO: This is a very easy function, where we choose the largest number Q value in Q matrix and find the best stragegy
:return:
'''
self.best_strategy = np.zeors(4,3)
def div_basis(self, bc):
mesh = self.mesh
p = self.p
divPhi = np.zeors((NC, ldof), dtype=self.dtype)
Dlambda, _ = mesh.grad_lambda()
Rlambda, _ = mesh.rot_lambda()
if p == 1:
divPhi[:, 0] = np.sum(Dlambda[:, 1, :] * Rlambda[:, 2, :] -
Dlambda[:, 2, :] * Rlambda[:, 1, :],
axia=1)
divPhi[:, 1] = np.sum(Dlambda[:, 1, :] * Rlambda[:, 2, :] +
Dlambda[:, 2, :] * Rlambda[:, 1, :],
axia=1)
divPhi[:, 2] = np.sum(Dlambda[:, 2, :] * Rlambda[:, 0, :] -
Dlambda[:, 0, :] * Rlambda[:, 2, :],
axis=1)
divPhi[:, 3] = np.sum(Dlambda[:, 2, :] * Rlambda[:, 0, :] +
Dlambda[:, 0, :] * Rlambda[:, 2, :],
axis=1)
divPhi[:, 4] = np.sum(Dlambda[:, 0, :] * Rlambda[:, 1, :] -
Dlambda[:, 1, :] * Rlambda[:, 0, :],
axis=1)
divPhi[:, 5] = np.sum(Dlambda[:, 0, :] * Rlambda[:, 1, :] +
Dlambda[:, 1, :] * Rlambda[:, 0, :],
axis=1)
divPhi[:, 0:6:2] *= cell2edgeSign
divPhi[:, 1:6:2] *= cell2edgeSign
else:
#TODO:raise a error
print("error")
return divPhi
def div_basis(self, bc):
mesh = self.mesh
p = self.p
divPhi = np.zeors((NC, ldof), dtype=self.dtype)
return divPhi
def div_basis(self, bc):
mesh = self.mesh
p = self.p
ldof = self.number_of_local_dofs()
NC = mesh.number_of_cells()
divPhi = np.zeors((NC, ldof), dtype=np.float)
cell2edgeSign = self.cell_to_edge_sign()
W = np.array([[0, 1], [-1, 0]], dtype=np.float)
Rlambda = mesh.rot_lambda()
Dlambda = Rlambda @ W
if p == 1:
divPhi[:, 0] = np.sum(
Dlambda[:, 1, :] * Rlambda[:, 2, :], axis=1) - np.sum(
Dlambda[:, 2, :] * Rlambda[:, 1, :], axis=1)
divPhi[:, 1] = np.sum(
Dlambda[:, 2, :] * Rlambda[:, 0, :], axis=1) - np.sum(
Dlambda[:, 0, :] * Rlambda[:, 2, :], axis=1)
divPhi[:, 2] = np.sum(
Dlambda[:, 0, :] * Rlambda[:, 1, :], axis=1) - np.sum(
Dlambda[:, 1, :] * Rlambda[:, 0, :], axis=1)
divPhi *= cell2edgeSign
else:
#TODO:raise a error
print("error")
return divPhi
def enkfanalysis(ensemble, obsvalue):
obs_error = 1.0E-7
obserrormat = np.mat([obs_error])
obsHmatrix = np.mat([0, 0, 1, 0, 0, 0])
matobsvalue = np.mat([obsvalue])
# mean variable
matensemble = np.mat(ensemble)
ensembleNum = len(ensemble[0])
meanvariable = np.mat(
[np.mean(ensemble[0, :]), np.mean(ensemble[1, :]), np.mean(ensemble[2, :]), np.mean(ensemble[3, :]),
np.mean(ensemble[4, :]), np.mean(ensemble[5, :])]).reshape((6, 1))
bh = np.mat((6, 1))
hbh = np.mat((1, 1))
for i in range(ensembleNum):
hx_hx = (np.mat(obsHmatrix) * np.mat(ensemble[:, i]) - np.mat(obsHmatrix) * meanvariable)
bh = (np.mat(ensemble[:, i]).reshape((6, 1)) - meanvariable) * (hx_hx.T) + bh # 6*1
hbh = hx_hx * (hx_hx.T) + hbh # 1*1
bh = bh / (ensembleNum - 1)
hbh = hbh / (ensembleNum - 1)
# kalman gain
kalmanGain = ph * ((hbh + obserrormat).I)
analysisensemble = np.mat(np.zeors((6, ensembleNum)))
for i in range(ensembleNum):
analysisensemble[:, i] = matensemble[:, i] + kalmanGain * (matobsvalue - obsHmatrix * matensemble[:, i])
return analysisensemble
def extract_features(imgs, feature_fns, verbose=False):
num_images = imgs.shape[0]
if num_images == 0:
return np.array([])
feature_dims = []
first_image_features = []
for feature_fn in feature_fns:
feats = feature_fn(imgs[0].squeeze())
assert len(
feats.shape) == 1, 'Feature functions must be one-dimensional'
feature_dims.append(feats.size)
first_image_features.append(feats)
total_feature_dim = sum(feature_dims)
imgs_features = np.zeors((num_images, total_feature_dim))
imgs_features[0] = np.hstack(first_image_features).T
for i in range(1, num_images):
idx = 0
for feature_fn, feature_dim in zip(feature_fns, feature_dims):
next_idx = idx + feature_dim
imgs_features[i, idx:next_idx] = feature_fn(imgs[i].squeeze())
idx = next_idx
if verbose and i % 1000 == 0:
print('Done extracting features for %d / $d images' %
(i, num_images))
return imgs_features
def p_jds(Z, L, U, C, M):
"""
Input:
coefficient matrix Z
desired numbers of dynamic active sets L
dictionary atom label vector U
the number of classes C
the number of views M
Output:
Index matrix I for top-L dynamic active sets
"""
#Initialize
I = np.zeros((L, M))
V = np.zeros((C, M))
_I = np.zeros((C, M))
S = np.zeors(C)
for l in xrange(L):
for i in xrange(C):
#c代表索引值
c = find(U, i)
for m in xrange(M):
v, t = Max(Z[c, m])
V[i, m] = v
_I[i, m] = c[t]
tmp = np.cumsum(np.power(V[i], 2))[-1]
S[i] = np.power(tmp, 0.5)
_v, _t = Max(S)
I[l, :] = _I[_t,:]
Z[_I[_t,:]] = 0
return I
def initialize(self):
'''
In this function we initialize our state space
'''
self.X = np.zeors(4, 4) # incoming cars
self.Y = np.zeros(4, 4) # waiting cars
self.Z = np.zeros(4, 4) # Outgoing cars
def onehot_vector(self, x, size):
if type(x) is not list:
onehot = np.zeros(size)
onehot[x] = 1
else:
onehot = np.zeors((len(x), size))
onehot[np.arange(len(x)), x] = 1
return onehot
def test_whole(self):
test_compound = np.zeors((self.compound_size, self.pos_compound_size +
self.neg_compound_size, self.compound_size))
test_gene = np.zeros(
(self.gene_size, self.pos_gene_size + self.neg_gene_size,
self.gene_size))
for i in range(self.compound_size):
compound[0, 0, :] = self.assign_value_compound(compoundid)
def crf(prob, area):
import pydensecrf.densecrf as dcrf
from pydensecrf.utils import unary_from_softmax, create_pairwise_bilateral
s_x = 16
s_y = 16
s_x_Gaussian = 4
s_y_Gaussian = 4
s_ch = 0.01
W = prob.shape[2]
H = prob.shape[1]
NLABELS = prob.shape[0]
image_path = ('top/top_mosaic_09cm_area%s.tif') % area
dsm_path = ('dsm/dsm_09cm_matching_area%s.tif') % area
image_rgb = imread(image_path)
image_dsm = imread(dsm_path)
image_rgb = image_rgb[:H, :W, :]
image_dsm = image_dsm[:H, :W]
#input_image = input_image.transpose(2,0,1)
U = unary_from_softmax(prob)
input_image_1 = image_rgb[:, :, 0:1]
pairwise_energy_1 = create_pairwise_bilateral(sdims=(s_x, s_y),
schan=(s_ch, ),
img=input_image_1,
chdim=2)
input_image_2 = image_rgb[:, :, 1:2]
pairwise_energy_2 = create_pairwise_bilateral(sdims=(s_x, s_y),
schan=(s_ch, ),
img=input_image_2,
chdim=2)
input_image_3 = image_rgb[:, :, 2:3]
pairwise_energy_3 = create_pairwise_bilateral(sdims=(s_x, s_y),
schan=(s_ch, ),
img=input_image_3,
chdim=2)
input_image_4 = image_dsm[:, :]
pairwise_energy_4 = create_pairwise_bilateral(sdims=(s_x, s_y),
schan=(s_ch, ),
img=input_image_4,
chdim=2)
d = dcrf.DenseCRF2D(W, H, NLABELS)
d.setUnaryEnergy(U)
d.addPairwiseEnergy(pairwise_energy_1, compat=10, kernel=dcrf.FULL_KERNEL)
d.addPairwiseEnergy(pairwise_energy_2, compat=10, kernel=dcrf.FULL_KERNEL)
d.addPairwiseEnergy(pairwise_energy_3, compat=10, kernel=dcrf.FULL_KERNEL)
d.addPairwiseEnergy(pairwise_energy_4, compat=10, kernel=dcrf.FULL_KERNEL)
d.addPairwiseGaussian(sxy=(s_x_Gaussian, s_y_Gaussian), compat=1)
Q = d.inference(5)
out_crf = np.argmax(Q, axis=0).reshape((H, W))
out_crf_expand = np.zeors(NLABELS, H, W)
for i in range(NLABELS):
out_crf_expand[i] = 1 * (out_crf == i)
return out_crf_expand
def test_load_data(path, sf_interest):
data = h5py.File(path, 'r')
gene_idx = data['gene_idx'][:]
ensg_ids = data['gene_names'][:]
conf_idx = data['conf_idx'][:]
interest_idx = _get_interest_idx(gene_idx, ensg_ids, sf_interest.values())
mask_idx = np.intersect1d(conf_idx, interest_idx)
mask = np.zeors(gene_idx.size, dtype=bool); mask[mask_idx] = True
return data, mask
def basis(self, bc):
mesh = self.mesh
dim = mesh.geom_dimension()
ldof = self.number_of_local_dofs()
p = self.p
phi = np.zeors((NC, ldof, dim), dtype=self.dtype)
return phi
def generate_output(t, a, sample_size=None):
if (t == 1):
if (type(sample_size) == type(0)):
return np.ones((sample_size, 1), dtype=np.float)
return np.ones((a.shape[0], 1), dtype=np.float)
elif (t == 0):
if (type(sample_size) == type(0)):
return np.zeros((sample_size, 1), dtype=np.float)
return np.zeors((a.shape[0], 1), dtype=np.float)
else:
print('argument t: {0, 1}')
def correlation(G,l,n,O):
corr=np.zeors(2)
for isite in range(0,2):
for i in range(n):
theta=np.tensordot(np.diag(l[(isite+i+1)%2,:]),G[(isite+i)%2,:,:,:],axes=(1,1))
theta=np.tensordot(theta,np.diag(l[(isite+i)%2,:]),axes=(2,0))
theta_o=np.tensordot(theta,O,axes=(1,0))
theta_o=np.tensordot(theta_o,O,axes=(-2,0)).conj()
oder_o=[0]+list(range(2,n+1))+[n+2]+[1,n+1]
corr[isite]=np.squeeze(np.tensordot(theta_o,theta,axes=(list(range(n+3)),oder_o))).item()
return corr
def kmeans1(boxes,k,dist=np.mean):
rows = boxes.shape[0]
distances = np.zeros([rows,k])
np.random.seed()
clusters = boxes[np.random.choice(rows, k, replace=False)]
cur_clusters = np.zeors([k,2])
last_clusters = np.zeors([rows,])
while(True):
for r in range(rows):
distances[r] = 1 - iou(boxes[r],clusters)
G = np.argmin(distances,axis = 1)
# 求均值
if (last_clusters== G).all():
break
for k_i in range(k):
cur_clusters[k_i] = dist(boxes[G==k_i],axis = 0)
clusters =cur_clusters
return clusters
def smiles2fpRdkit(smiles, explicit=128):
if len(smiles) == 0:
fp = '\n'
print('Warning by RM: empty smiles!')
arr = np.zeors((1, explicit), dtype=int)
else:
ms = Chem.MolFromSmiles(smiles)
fp = AllChem.GetMorganFingerprintAsBitVect(ms, 2, nBits=explicit)
arr = np.zeros((1, ), dtype=int)
DataStructs.ConvertToNumpyArray(fp, arr)
return arr
def getSampleCovariance(data, mean):
sampleCovariance = np.zeors((Data.D, Data.D))
for i in range(Data.numVectors):
x_minus_x_bar = np.zeros((Data.D, 1))
x_minus_x_bar = data[i] + x_minus_x_bar
x_minus_x_bar = x_minus_x_bar - mean
x_minus_x_bar_T = x_minus_x_bar.T
mul = np.dot(x_minus_x_bar, x_minus_x_bar_T)
sampleCovariance += mul
sampleCovariance /= Data.numVectors - 1
return sampleCovariance
def var_estimation(X):
'''
Use superimposed method to compute multi-dimension data's variance
:param X: input data array
'''
if X.ndim == 3: # (n_events, n_epochs, n_times)
var = np.zeors((X.shape[0], X.shape[2])) # (n_events, n_times)
for i in range(X.shape[0]): # i for n_events
ex = np.mat(np.mean(X[i, :, :], axis=0)) # (1, n_times)
temp = np.mat(np.ones((1, X.shape[1]))) # (1, n_epochs)
minus = (temp.T * ex).A # (n_epochs, n_times)
var[i, :] = np.mean((X[i, :, :] - minus)**2, axis=0)
elif X.ndim == 2: # (n_epochs, n_times)
var = np.zeors((X.shape[1])) # (n_times)
ex = np.mat(np.mean(X, axis=0)) # (1, n_times)
temp = np.mat(np.ones(1, X.shape[0])) # (1, n_epochs)
minus = (temp.T * ex).A # (n_epochs, n_times)
var = np.mean((X - minus)**2, axis=0)
return var
def getSubstack(stack1, r, sgX, sgY, sgZ):
sX, sY, sZ = stack1.shape
substack = np.zeors([sgX, sgY, sgZ])
# int ii,jj,kk
sum = 0
# int iii,jjj,kkk
# std::cout<<xf<<","<<yf<<","<<zf<<","<<dxf<<","<<dyf<<","<<dzf<<" | "<<sX<<","<<sY<<","<<sZ<<" \n"
xf, yf, zf = r.astype(int)
fdi = r[0] - xf
fdj = r[1] - yf
fdk = r[2] - zf
if (sgX / 2) < xf < (sX - sgX / 2) and (sgY / 2) < yf < (
sY - sgY / 2) and (sgZ / 2) < zf < (sZ - sgZ / 2):
# std::cout<<xf<<","<<yf<<","<<zf<<", | "<<sX<<","<<sY<<","<<sZ<<" \n"
x0 = xf - sgX // 2
x1 = x0 - sgX
y0 = yf - sgY // 2
y1 = y0 - sgY
z0 = zf - sgZ // 2
z1 = z0 - sgZ
substack = (
(1 - fdi) * (1 - fdj) * (1 - fdk) * stack1[x0:x1, y0:y1, z0:z1] +
(fdi) * (1 - fdj) *
(1 - fdk) * stack1[x0 + 1:x1 + 1, y0:y1, z0:z1] + (1 - fdi) *
(fdj) * (1 - fdk) * stack1[x0:x1, y0 + 1:y1, z0:z1] + (1 - fdi) *
(1 - fdj) * (fdk) * stack1[x0:x1, y0:y1, z0 + 1:z1] + (fdi) *
(fdj) * (1 - fdk) * stack1[x0 + 1:x1 + 1, y0 + 1:y1 + 1, z0:z1] +
(1 - fdi) * (fdj) *
(fdk) * stack1[x0:x1, y0 + 1:y1 + 1, z0 + 1:z1 + 1] + (fdi) *
(1 - fdj) * (fdk) * stack1[x0 + 1:x1 + 1, y0:y1, z0 + 1:z1 + 1] +
(fdi) * (fdj) *
(fdk) * stack1[x0 + 1:x1 + 1, y0 + 1:y1 + 1, z0 + 1:z1 + 1:])
substack -= np.mean(substack)
substack /= np.linalg.norm(substack)
return substack
else:
substack[:] = 0
return substack
def hessian(self,x):
n = x.shape[0]
if self.method =="NPS" || self.method ==1:
w=self.w;
xi = self.xi;
N = x.shape[1]
H = np.zeors((n))
for ii in range(N):
X = x - xi[:,ii]
if np.linalg.norm(X) > 1e-5:
H = H + 3*w[ii]*((X*X.T)/np.linalg.norm(X) + np.linalg.norm(X)*np.identity(n))
return H
def run(self):
while True:
self.predict()
policy_stable = True
for state_i in range(self.state_values_func.size()):
old_action = copy.deepcopy(self.policy[state_i])
state_action_value = np.zeros(self.env.action_space.n)
for action_i in self.env.action_space:
self.env.set_current_state(state_i)
next_state, reward, _, _ = self.env.step(action_i)
state_action_value[
action_i] = reward + self.gamma * self.state_values_func[
next_state]
optimal_action = np.random.choice(
np.flatnonzero(
state_action_value == state_action_value.max()))
np.zeors(self.policy[state_i])
self.policy[state_i][optimal_action] = 1.0
if old_action != self.policy:
policy_stable = False
if policy_stable:
return
def GenralBrownianMotion(T, mu= lambda t, Wt: 0,
sigma= lambda t, Wt: 1,
B0 = 1, steps=1000,
seed=rand_seed()):
dW = BrownianPath(T, steps, seed)
W = np.zeors((steps+1,))
dt = T/steps
T = [i*dt for i in range(steps+1)]
for idx, dw in enumerate(dW):
t=T[idx]
w=W[idx]
_mu = mu(t,w)
_sigma = sigma(t,w)
W[idx+1] = _mu*dt + _simga*dw
_W = UnitLocalLinearCurveFromArray(zip(T,W))
return _W
def predict(self, X):
""" X in N x D where each row is an example we wish to predict label for """
num_test = X.shape[0]
# lets make sure that the output type mathces the input type
Ypred = np.zeors(num_test, dtype=self.ytr.dtype)
# loop over all test rows
for i in xrange(num_test):
# find the nereat training image to the i'th thest image
# using the L1 distance (sum of absolute value differences)
distances = np.sum(np.abs(self.Xtr=X[i, :]), axis=1)
min_index = np.argmin(
distances) # get the index with smallest distance
Ypred[i] = self.ytr[min_index]
return Ypred
def get_overview(code, headers):
url = f'http://comp.fnguide.com/SVO2/ASP/SVD_Main.asp?pGB=1&gicode=A{code}&cID=&MenuYn=Y&ReportGB=&NewMenuID=101&stkGb=701'
req = requests.get(url, headers=headers)
soup = BeautifulSoup(req.content, 'html.parser')
table = soup.select('#svdMainGrid1 > table')
_state = pd.read_html(str(table))[0]
col = ['전일종가', '시가총액', '발행주식수_보통주', '발행주식수_우선주']
try:
price = int(_state.loc[0][1].split('/')[0].replace(',', ''))
Market_cap = int(_state.loc[4][1])
shares_norm = int(_state.loc[6][1].split('/')[0].replace(',', ''))
shares_pre = int(_state.loc[6][1].split('/')[1].replace(',', ''))
return col, [price, Market_cap, shares_norm, shares_pre]
except:
return col, np.zeors(len(col))
def ori_img(self, input, x, y, reflect_i):
r = 3 * self.sigma_s
a = np.arange(-r, r + 1)
if len(input.shape) is 2:
input_diff = np.zeors((len(a), len(a), 1))
else:
input_diff = np.zeros((len(a), len(a), input.shape[2]))
for j in range(len(a)):
for i in range(len(a)):
input_diff[i][j] = np.float64(reflect_i[x - a[i] +
r][y - a[j] + r])
return input_diff
def div_basis(self, bc):
mesh = self.mesh
p = self.p
divPhi = np.zeors((NC, ldof), dtype=self.dtype)
cell2edgeSign = self.cell_to_edge_sign()
Dlambda, _ = mesh.grad_lambda()
if p == 0:
divPhi[:, 0] = np.sum(Dlambda[:, 1, :]*Dlambda[:, 2, :] - Dlambda[:, 2, :]*Dlambda[:, 1, :], axia=1)
divPhi[:, 1] = np.sum(Dlambda[:, 2, :]*Dlambda[:, 0, :] - Dlambda[:, 0, :]*Dlambda[:, 2, :], axis=1)
divPhi[:, 2] = np.sum(Dlambda[:, 0, :]*Dlambda[:, 1, :] - Dlambda[:, 1, :]*Dlambda[:, 0, :], axis=1)
divPhi *= cell2edgeSign
else:
#TODO:raise a error
print("error")
return divPhi