def test_RelativePoseFromTwoPointsWithKnownRotation(R0, R1, pts0, pts1, gt_pose):
unrotated_norm_img_pts0 = []
unrotated_norm_img_pts1 = []
for p0, p1 in zip(img_pts0, img_pts1):
unrotated_norm_img_pts0.append(R0.T @ np.extend(p0, 1))
unrotated_norm_img_pts1.append(R1.T @ np.extend(p1, 1))
success, position2 = pt.sfm.RelativePoseFromTwoPointsWithKnownRotation(
unrotated_norm_img_pts0, unrotated_norm_img_pts1)
def nextPagePlusOne(self, response, url):
np = []
nextPageText = ''.join(response.xpath(self.xpath['nextPageText']).extract()).strip()
if nextPageText == '下一页':
np.extend(response.xpath(self.xpath['nextPage']).extract())
else:
p = response.xpath(self.xpath['allPage'])
# 框架支持url排重,这里就不排重了
for one in p:
np.extend(url + one.xpath('.//@href').extract())
return np
def nextPage(self, response):
np = []
nextPageText = ''.join(response.xpath(self.xpath['nextPageText']).extract()).strip()
if nextPageText == '下一页':
np.extend(response.xpath(self.xpath['nextPage']).extract())
else:
p = response.xpath(self.xpath['allPage'])
# 框架支持url排重,这里就不排重了
for one in p:
np.extend(one.xpath('.//@href').extract())
if len(np) == 0:
pass
return np
def update(self, alpha, G, s_cont, s_cate, a_tau):
s_tau = np.extend(self.normalize(s_cont), s_cate)
X = np.reshape(np.append(s_tau, [a_tau]), (1, self.input_dims))
Y = np.reshape(np.array([G]), (
1,
1,
))
self.sess.run(self.train_op, feed_dict={self.X: X, self.Y: Y})
def feedForward(x, Theta_list, depth):
Theta = Theta_list[0]
# appending column vector on ones to the beginning of the input matrix
x = np.extend(np.ones((x.shape[0], 1)), x)
##a = np.ones((Theta.shape[0],1))
# iteration variable
a = sigmoid(x.dot(np.transpose(Theta)))
if depth > 0:
del (Theta_list[0])
return feedForward(a, Theta_list, depth - 1)
elif depth == 0:
return a
else:
print("Depth cannot be less than 0")
def plot_confusion_matrix(self, rel_true, rel_pred, filename="_ConfusionMatrix.png", cmap=plt.cm.Blues):
confusion = confusion_matrix(rel_true, rel_pred)
uniqRel = np.unique(rel_true)
uniqRel = np.unique(np.extend(uniqRel,rel_pred))
fig = plt.figure(figsize=(8,8))
ax = fig.add_subplot(111)
image = ax.matshow(confusion, cmap=cmap) # imshow
fig.colorbar(image)
tick_marks = np.arange(confusion.shape[0])
ax.set_xlabel("Predicted")
ax.set_xticks(tick_marks)
ax.set_xticklabels(uniqRel)
ax.set_ylabel("Actual")
ax.set_yticks(tick_marks)
ax.set_yticklabels(uniqRel)
plt.setp(ax.xaxis.get_majorticklabels(), rotation=45)
plt.savefig(self.outfolder+self.tag+filename, bbox_inches='tight')
def __call__(self, s_cont, s_cate, a):
# TODO: implement this method
s = np.extend(self.normalize(s_cont), s_cate)
X = np.reshape(np.append(s, [a]), (1, self.input_dims))
pred = self.sess.run(self.Y_hat, feed_dict={self.X: X})
return pred[0, 0]
def loadtxt(filename,
infer_format=True,
directed=None,
weighted=False,
sep=' ',
save=False):
"""
Loads a text file into a sparse graph. Assumes a 2 or 3 column format with
columns separated by sep. The first two columns should specify an edge from
the first to the second. If directed is False, this will be entered as an
undirected edge between the first and the second. The third column is optional,
and should specify the weight of the edge.
NOTES:
DOES NOT SUPPORT MULTIGRAPHS. If a multigraph is passed in and weighted
is true, the duplicates will be summed together into a single connection.
In the case that weighted is False, duplicate entries are ignored and
the standard adjacency matrix will be created.
REMOVES ALL SELF EDGES. Our specialization code cannot handle that yet.
Parameters:
filename (str): path to text file
infer_format (bool): If True (default), attempts to automatically infer
the format. Even when infer_format=True, other keyword arguments can
be used to specify certain behaviors (see below).
directed (bool): If None (default), infer_format must be true; in this
case, attempts to infer if the graph is directed or not. If True/False,
graph will be treated as directed/undirected, regardless of if
infer_format is True of False.
weighted (bool): If False (default), graph is constructed as unweighted,
regardless of whether there are weights or not in the file. If True,
weights are used. Raises a ValueError if True but no weights present.
sep (str): string to split columns by
save (bool/str): If False (default), does nothing. If a string specifying
a save path is given, the csr_matrix is pickled at that location.
Returns:
G (sparse.csr_matrix): sparse matrix of graph data
"""
# dictionaries mapping [un]directed/[un]weighted format strings to bools
dir_dict = {'sym': False, 'asym': True}
weight_dict = {'unweighted': False, 'posweighted': True, 'signed': True}
# load file
with open(filename, 'r') as f:
data = f.read()
data = data.split('\n')
format_str = data[0].split(sep)[
1:] # first line of document is format string
# sometimes there are other lines with info that isn't needed
rm_ind = 0
for elem in range(len(data)):
# run through the lines of data until we find the first used line
if data[elem][0] == '%':
rm_ind = elem + 1
else:
break
data = data[rm_ind:-1] # last line is empty
if infer_format:
# try to infer sep
if '\t' in data[0]:
sep = '\t'
elif ',' in data[0]:
sep = ','
# if it isn't this, then it is likely ' ' (default), or user specified
# split into list of lists; inner lists contain connection data
data = [entry.strip().split(sep) for entry in data]
data = np.array(data,
dtype=np.int) # numpyify and cast the strings to ints
if infer_format:
try:
# does the document tell us if this is weighted or not?
weight_format = weight_dict[format_str[1]]
except KeyError:
if weighted is False:
# if the user doesn't care about weights, forget about it
pass
else:
# otherwise, yell at them for doing something confusing
print(('KeyError: "{}" is an unknown formatting option for '
'weighted/unweighted.'.format(format_str[1])))
print('Current known options are: {}.'.format(weight_dict))
return None
# number of columns specifies the formatting
rows, cols = data.shape
# weighted inference
if cols == 2 and weighted:
# if user specified weighted, but there are no weights
raise ValueError(
f'File "{filename}" has no weights, but weighted was set to True.'
)
elif cols == 4 and weighted:
# if users specified weighted, but we have the weird 4 column format
if np.all(data[:, 2] == 1):
# if the third column is ones, fourth column is weights
data = data[:, [0, 1, 3]] # remove column of ones
else:
# if third column not just ones, we don't know what to do
raise ValueError('Weights specified, but file format unknown')
elif cols >= 3 and weighted is False:
# if we have weights, but the user said to ignore them
data = data[:, :2]
# directed inference
if directed is None:
try:
directed = dir_dict[format_str[0]]
except KeyError:
print(('KeyError: "{}" is an unknown formatting option for '
'directed/undirected.'.format(format_str[0])))
print('Current known options are: {}.'.format(format_str[0]))
# ------------------------------------------------------------------------ #
# at this point we should have directed and weighted properly specified
# explicitly extract row and column indices
col_inds, row_inds = data[:, 0], data[:, 1]
# note our convention that G[i, j]=1 means that node[i] recieves from node[j]
# get proper shape
shape = max(np.max(row_inds), np.max(col_inds)) + 1
shape = (shape, shape)
# construct the sparse graph based on directed/undirected
if directed:
if weighted:
# construct weighted, directed graph
G = sparse.coo_matrix((data[:, -1], (row_inds, col_inds)),
shape=shape)
# NOTE THAT THIS SUMS DUPLICATE EDGES
else:
# we don't want duplicates to be summed
unique_edges = {(i, j) for (i, j) in zip(row_inds, col_inds)}
data = np.array([[i, j] for (i, j) in unique_edges])
# row_inds/col_inds extraction reverse from above b/c we already
# accounted for the row <-- col structure
row_inds, col_inds = data[:, 0], data[:, 1]
rows = row_inds.size
# construct unweighted, directed graph
G = sparse.coo_matrix((np.ones(rows), (row_inds, col_inds)),
shape=shape)
else:
# if the graph is undirected, we need to add both i <--> j connections
# extend row/col inds with col/row inds
row_inds, col_inds = np.append(row_inds, col_inds), np.append(
col_inds, row_inds)
rows = row_inds.size
if weighted:
# extend weights
weights = np.extend(data[:, 2], data[:, 2])
# construct weighted, undirected graph
G = sparse.coo_matrix((weights, (row_inds, col_inds)), shape=shape)
# NOTE THAT THIS SUMS DUPLICATE EDGES
else:
# remove duplicate edges
unique_edges = {(i, j) for (i, j) in zip(row_inds, col_inds)}
data = np.array([[i, j] for (i, j) in unique_edges])
# row_inds/col_inds extraction reverse from above b/c we already
# accounted for the row <-- col structure
row_inds, col_inds = data[:, 0], data[:, 1]
rows = row_inds.size
# construct unweighted, undirected graph
G = sparse.coo_matrix((np.ones(rows), (row_inds, col_inds)),
shape=shape)
# REMOVE SELF EDGES, WE CANNOT HANDLE THEM YET
G.setdiag(0)
G.tocsr()
if save:
# pickle the graph
try:
with open(save, 'wb') as f:
pickle.dump(G, f)
except FileNotFoundError as e:
# if specified path doesn't exist, print the error, then return G
print(e)
return G
return G
label = numpy.zeros(max_nr_class, dtype = int) count = numpy.zeros(max_nr_class, dtype = int) for i in range(l): this_label = int(prob.y[i]) j = 0 for k in range(nr_class): j = k if this_label is label[j]: count[j] += 1 break if j is nr_class: if nr_class is max_nr_class: max_nr_class *= 2 zero = numpy.zeros(max_nr_class, dtype = int) label = numpy.extend(label, zero) count = numpy.extend(label, zero) label[nr_class] = this_label count[nr_class] = 1 nr_class += 1 for i in range(nr_class): n1 = count[i] for j in range(i+1, nr_class): n2 = count[j] if param.nu * (n1 + n2)/2.0 > min(n1,n2): return "specifies nu is infeasible" else: return None def svm_check_probability_model(self, model):
