How to use norm_rows method in Testify

Best Python code snippet using Testify_python

test_np_distances.py

Source:test_np_distances.py

1from gehm.utils.np_distances import *2from tests.test_data import create_test_data34import pytest5import numpy as np6import torch7from numpy import cos, sin89@pytest.mark.distances10def test_nx_second_order_proximity(create_test_data):1112    G,G_undir=create_test_data13    nodes=np.array(G.nodes)141516    # Test 1: Ordering when subset proximity17    sel1=nodes[[0,1,2,3]]18    sel2=nodes[[0,3,2,1]]19    prox1=nx_second_order_proximity(G,sel1,False)20    prox2=nx_second_order_proximity(G,sel2,False)21    prox1=np.round(prox1,5)22    prox2=np.round(prox2,5)23    assert (prox1[0,1]==prox2[0,3]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])24    assert (prox1[1,2]==prox2[3,2]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])252627    # Test 2: Ordering when whole network proximity28    prox1=nx_second_order_proximity(G,sel1,True)29    prox2=nx_second_order_proximity(G,sel2,True)30    prox1=np.round(prox1,5)31    prox2=np.round(prox2,5)32    assert (prox1[1,3]==prox2[3,3]), "Ordering problem, {} != {}".format(prox1[1,3],prox2[3,3])33    assert (prox1[3,1]==prox2[1,1]), "Ordering problem, {} != {}".format(prox1[3,1],prox2[1,1])3435    # Test 3+4: Without row normalization36    prox1=nx_second_order_proximity(G,sel1,False, norm_rows=False)37    prox2=nx_second_order_proximity(G,sel2,False, norm_rows=False)38    prox1=np.round(prox1,5)39    prox2=np.round(prox2,5)40    assert (prox1[0,1]==prox2[0,3]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])41    assert (prox1[1,2]==prox2[3,2]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])4243    prox1=nx_second_order_proximity(G,sel1,True, norm_rows=False)44    prox2=nx_second_order_proximity(G,sel2,True, norm_rows=False)45    prox1=np.round(prox1,5)46    prox2=np.round(prox2,5)47    assert (prox1[1,3]==prox2[3,3]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])48    assert (prox1[3,1]==prox2[1,1]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])4950    # Test 5+6: Without row normalization, but with batch normalization51    prox1=nx_second_order_proximity(G,sel1,False, norm_rows=False, norm_rows_in_sample=True)52    prox2=nx_second_order_proximity(G,sel2,False, norm_rows=False, norm_rows_in_sample=True)53    prox1=np.round(prox1,5)54    prox2=np.round(prox2,5)55    assert (prox1[0,1]==prox2[0,3]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])56    assert (prox1[1,2]==prox2[3,2]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])5758    prox1=nx_second_order_proximity(G,sel1,True, norm_rows=False, norm_rows_in_sample=True)59    prox2=nx_second_order_proximity(G,sel2,True, norm_rows=False, norm_rows_in_sample=True)60    prox1=np.round(prox1,5)61    prox2=np.round(prox2,5)62    assert (prox1[1,3]==prox2[3,3]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])63    assert (prox1[3,1]==prox2[1,1]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])646566    # Test 7: Whole network, but return batch order67    prox1=nx_second_order_proximity(G,sel1,True, norm_rows=True, to_batch=True)68    prox2=nx_second_order_proximity(G,sel2,True, norm_rows=True, to_batch=True)69    prox1=np.round(prox1,5)70    prox2=np.round(prox2,5)71    assert (prox1[1,0]==prox2[3,0]), "Ordering problem, {} != {}".format(prox1[1,0],prox2[3,0])72    assert (prox1[3,1]==prox2[1,3]), "Ordering problem, {} != {}".format(prox1[3,1],prox2[1,3])7374    # Test 8: Whole network, but return batch order, no norm75    prox1=nx_second_order_proximity(G,sel1,True, norm_rows=False, to_batch=True)76    prox2=nx_second_order_proximity(G,sel2,True, norm_rows=False, to_batch=True)77    prox1=np.round(prox1,5)78    prox2=np.round(prox2,5)79    assert (prox1[1,0]==prox2[3,0]), "Ordering problem, {} != {}".format(prox1[1,0],prox2[3,0])80    assert (prox1[3,1]==prox2[1,3]), "Ordering problem, {} != {}".format(prox1[3,1],prox2[1,3])81828384    # Now repeat everything with an undirected graph:8586    G=G_undir87    nodes=np.array(G.nodes)888990    # Test 1: Ordering when subset proximity91    sel1=nodes[[0,1,2,3]]92    sel2=nodes[[0,3,2,1]]93    prox1=nx_second_order_proximity(G,sel1,False)94    prox2=nx_second_order_proximity(G,sel2,False)95    prox1=np.round(prox1,5)96    prox2=np.round(prox2,5)97    assert (prox1[0,1]==prox2[0,3]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])98    assert (prox1[1,2]==prox2[3,2]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])99100101    # Test 2: Ordering when whole network proximity102    prox1=nx_second_order_proximity(G,sel1,True)103    prox2=nx_second_order_proximity(G,sel2,True)104    prox1=np.round(prox1,5)105    prox2=np.round(prox2,5)106    assert (prox1[1,3]==prox2[3,3]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])107    assert (prox1[3,1]==prox2[1,1]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])108109    # Test 3+4: Without row normalization110    prox1=nx_second_order_proximity(G,sel1,False, norm_rows=False)111    prox2=nx_second_order_proximity(G,sel2,False, norm_rows=False)112    prox1=np.round(prox1,5)113    prox2=np.round(prox2,5)114    assert (prox1[0,1]==prox2[0,3]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])115    assert (prox1[1,2]==prox2[3,2]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])116117    prox1=nx_second_order_proximity(G,sel1,True, norm_rows=False)118    prox2=nx_second_order_proximity(G,sel2,True, norm_rows=False)119    prox1=np.round(prox1,5)120    prox2=np.round(prox2,5)121    assert (prox1[1,3]==prox2[3,3]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])122    assert (prox1[3,1]==prox2[1,1]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])123124    # Test 5+6: Without row normalization, but with batch normalization125    prox1=nx_second_order_proximity(G,sel1,False, norm_rows=False, norm_rows_in_sample=True)126    prox2=nx_second_order_proximity(G,sel2,False, norm_rows=False, norm_rows_in_sample=True)127    prox1=np.round(prox1,5)128    prox2=np.round(prox2,5)129    assert (prox1[0,1]==prox2[0,3]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])130    assert (prox1[1,2]==prox2[3,2]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])131132    prox1=nx_second_order_proximity(G,sel1,True, norm_rows=False, norm_rows_in_sample=True)133    prox2=nx_second_order_proximity(G,sel2,True, norm_rows=False, norm_rows_in_sample=True)134    prox1=np.round(prox1,5)135    prox2=np.round(prox2,5)136    assert (prox1[1,3]==prox2[3,3]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])137    assert (prox1[3,1]==prox2[1,1]), "Ordering problem, {} != {}".format(prox1[0,1],prox2[0,3])138139140    # Test 7: Whole network, but return batch order141    prox1=nx_second_order_proximity(G,sel1,True, norm_rows=True, to_batch=True)142    prox2=nx_second_order_proximity(G,sel2,True, norm_rows=True, to_batch=True)143    prox1=np.round(prox1,5)144    prox2=np.round(prox2,5)145    assert (prox1[1,0]==prox2[3,0]), "Ordering problem, {} != {}".format(prox1[1,0],prox2[3,0])146    assert (prox1[3,1]==prox2[1,3]), "Ordering problem, {} != {}".format(prox1[3,1],prox2[1,3])147148    # Test 8: Whole network, but return batch order, no norm149    prox1=nx_second_order_proximity(G,sel1,True, norm_rows=False, to_batch=True)150    prox2=nx_second_order_proximity(G,sel2,True, norm_rows=False, to_batch=True)151    prox1=np.round(prox1,5)152    prox2=np.round(prox2,5)153    assert (prox1[1,0]==prox2[3,0]), "Ordering problem, {} != {}".format(prox1[1,0],prox2[3,0])154    assert (prox1[3,1]==prox2[1,3]), "Ordering problem, {} != {}".format(prox1[3,1],prox2[1,3])
...

cosine.py

Source:cosine.py

1from ..utils import _XXT, rowSum, reflect, setDiagonal2from numpy import zeros, newaxis3from numba import njit, prange4from ..sparse.csr import div_1 ,mult_1, _XXT as _XXT_sparse, rowSum as rowSum_sparse5from ..sparse.tcsr import _XXT as _XXT_triangular6def cosine_sim_triangular(N, D):7	"""8	[Added 21/10/2018]9	Quickly performs X / norm_rows / norm_rows.T on the TCSR matrix.10	"""11	n = len(N)12	move = 013	14	# loop *-2 and adds S[:, newaxis]15	for i in prange(n-1):16		i1 = i+117		18		left = i*i1 // 219		s = N[i1]20		for j in range(left, left+i1):			21			# div N[:, newaxis]22			D[j] /= s23	24	# loop div N[newaxis, :]25	for a in prange(n-1):26		s = N[a]27		for b in range(a, n-1):28			# div N[newaxis, :] or N29			c = b*(b+1) // 2 + a30			D[c] /= s31	return D32cosine_sim_triangular_single = njit(cosine_sim_triangular, fastmath = True, nogil = True, cache = True)33cosine_sim_triangular_parallel = njit(cosine_sim_triangular, fastmath = True, nogil = True, parallel = True)34@njit(fastmath = True, nogil = True, cache = True)35def cosine_dis(XXT):36	"""37	[Added 22/10/2018]38	Performs XXT*-1 + 1 quickly on the lower triangular part.39	"""40	n = len(XXT)41	for i in range(n):42		for j in range(i):43			XXT[i, j] *= -144			XXT[i, j] += 145	return XXT46@njit(fastmath = True, nogil = True, cache = True)47def cosine_dis_triangular(D):48	"""49	[Added 22/10/2018]50	Performs XXT*-1 + 1 quickly on the TCSR.51	"""52	D *= -153	D += 154	return D55def cosine_similarity(X, Y = None, triangular = False, n_jobs = 1, copy = False):56	"""57	[Added 20/10/2018] [Edited 22/201/2018]58	[Edited 22/10/2018 Added Y option]59	Note: when using Y, speed improvement is approx 5% only from Sklearn.60	Cosine similarity is approx the same speed as Sklearn, but uses approx 10%61	less memory. One clear advantage is if you set triangular to TRUE, then it's faster.62	"""63	norm_rows = rowSum(X, norm = True)64	if Y is X:65		# Force algo to be triangular cosine rather than normal CS.66		Y = None67	if Y is None:68		if copy:69			XXT = _XXT(X.T)70			XXT /= norm_rows[:, newaxis]71			XXT /= norm_rows #[newaxis, :]72		else:73			XXT = _XXT(  (X/norm_rows[:, newaxis]).T  )74		if not triangular:75			XXT = reflect(XXT, n_jobs)76		# diagonal is set to 177		setDiagonal(XXT, 1)78		return XXT79	else:80		D = X @ Y.T81		D /= norm_rows[:, newaxis]82		D /= rowSum(Y, norm = True)83		return D84def cosine_similarity_sparse(val, colPointer, rowIndices, n, p, triangular = False, dense_output = True,85	n_jobs = 1, copy = True):86	"""87	[Added 20/10/2018] [Edited 21/10/2018]88	Slightly faster than Sklearn's Cosine Similarity implementation.89	If dense_output is set to FALSE, then a TCSR Matrix (Triangular CSR Matrix) is90	provided and not a CSR matrix. This has the advantage of using only 1/2n^2 - n91	memory and not n^2 memory.92	"""93	norm_rows = rowSum_sparse(val, colPointer, rowIndices, norm = True)94	if dense_output:95		if copy:96			XXT = _XXT_sparse(val, colPointer, rowIndices, n, p, n_jobs)97			XXT /= norm_rows[:, newaxis]98			XXT /= norm_rows #[newaxis, :]99		else:100			val = div_1(val, colPointer, rowIndices, norm_rows, n, p, copy = False)101			XXT = _XXT_sparse(val, colPointer, rowIndices, n, p, n_jobs)102			val = mult_1(val, colPointer, rowIndices, norm_rows, n, p, copy = False)103		if not triangular: 104			XXT = reflect(XXT, n_jobs)105		# diagonal is set to 1106		setDiagonal(XXT, 1)107	else:108		XXT = _XXT_triangular(val, colPointer, rowIndices, n, p, n_jobs)109		XXT = cosine_triangular_parallel(norm_rows, XXT) if n_jobs != 1 else \110			cosine_triangular_single(norm_rows, XXT)111	return XXT112def cosine_distances(X, Y = None, triangular = False, n_jobs = 1, copy = False):113	"""114	[Added 15/10/2018] [Edited 18/10/2018]115	[Edited 22/10/2018 Added Y option]116	Note: when using Y, speed improvement is approx 5-10% only from Sklearn.117	Slightly faster than Sklearn's Cosine Distances implementation.118	If you set triangular to TRUE, the result is much much faster.119	(Approx 50% faster than Sklearn)120	"""121	norm_rows = rowSum(X, norm = True)122	if Y is X:123		# Force algo to be triangular cosine rather than normal CS.124		Y = None125	if Y is None:126		if copy:127			XXT = _XXT(X.T)128			XXT /= norm_rows[:, newaxis]129			XXT /= norm_rows #[newaxis, :]130		else:131			XXT = _XXT(  (X/norm_rows[:, newaxis]).T  )132		# XXT*-1 + 1133		XXT = cosine_dis(XXT)134		if not triangular:135			XXT = reflect(XXT, n_jobs)136		# diagonal is set to 0 as zero distance between row i and i137		setDiagonal(XXT, 0)138		return XXT139	else:140		D = X @ Y.T141		D /= norm_rows[:, newaxis]142		D /= rowSum(Y, norm = True)143		D *= -1144		D += 1145		return D146def cosine_distances_sparse(val, colPointer, rowIndices, n, p, triangular = False, dense_output = True,147	n_jobs = 1, copy = True):148	"""149	[Added 22/10/2018]150	Slightly faster than Sklearn's Cosine Distances implementation.151	If dense_output is set to FALSE, then a TCSR Matrix (Triangular CSR Matrix) is152	provided and not a CSR matrix. This has the advantage of using only 1/2n^2 - n153	memory and not n^2 memory.154	"""155	norm_rows = rowSum_sparse(val, colPointer, rowIndices, norm = True)156	if dense_output:157		if copy:158			XXT = _XXT_sparse(val, colPointer, rowIndices, n, p, n_jobs)159			XXT /= norm_rows[:, newaxis]160			XXT /= norm_rows #[newaxis, :]161		else:162			val = div_1(val, colPointer, rowIndices, norm_rows, n, p, copy = False)163			XXT = _XXT_sparse(val, colPointer, rowIndices, n, p, n_jobs)164			val = mult_1(val, colPointer, rowIndices, norm_rows, n, p, copy = False)165		# XXT*-1 + 1166		XXT = cosine_dis(XXT)167		if not triangular: 168			XXT = reflect(XXT, n_jobs)169		# diagonal is set to 0 as zero distance between row i and i170		setDiagonal(XXT, 0)171	else:172		XXT = _XXT_triangular(val, colPointer, rowIndices, n, p, n_jobs)173		# XXT*-1 + 1174		XXT = cosine_dis_triangular(XXT)175		XXT = cosine_triangular_parallel(norm_rows, XXT) if n_jobs != 1 else \176			cosine_triangular_single(norm_rows, XXT)...

distances.py

Source:distances.py

1import networkx as nx2from typing import Union3from torch import Tensor4import networkx as nx5import torch6from torch.nn.functional import cosine_similarity7from torch import cdist8from gehm.utils.funcs import row_norm91011def embedding_first_order_proximity(12    positions: Tensor, norm_rows: bool = True,13) -> Tensor:14    """15    A simple application of euclidian distance to the positions vector in the embedding space.16    Includes row normalization to get relative distances.1718    Parameters19    ----------20    positions : Tensor21        Input positions, usually nx222    norm_rows : bool, optional23        If True, rows will be normed to 1, by default True2425    Returns26    -------27    Tensor28        Similarity Matrix29    """3031    assert isinstance(positions, Tensor)3233    similarity_matrix = cdist(positions, positions, p=2).to(positions.device)34    if norm_rows:35        similarity_matrix = row_norm(similarity_matrix)36    similarity_matrix = 1-similarity_matrix373839    return similarity_matrix-torch.eye(similarity_matrix.shape[0]).to(positions.device)404142def embedding_second_order_proximity(43    positions: Tensor, norm_rows: bool = True,44) -> Tensor:45    """46    Derives first the pairwise distances, then compares these distance vectors47    between positions to derive proximities of second order.4849    Parameters50    ----------51    positions : Union[Tensor,ndarray]52        Input positions, usually nx253    norm_rows : bool, optional54        If True, rows will be normed to 1, by default True5556    Returns57    -------58    Tensor59        Similarity Matrix60    """61    similarity_matrix = embedding_first_order_proximity(positions, norm_rows=norm_rows)62    similarity_matrix = matrix_cosine(similarity_matrix)6364    if norm_rows:65        similarity_matrix = row_norm(similarity_matrix)66    return similarity_matrix.to(positions.device)676869def matrix_cosine(mat: Tensor) -> Tensor:70    """71    Applies cosine similarity to a matrix, since pdist or cdist in PyTorch72    only computes Minkowski metrices.7374    Parameters75    ----------76    mat : Tensor77        NxN Input7879    Returns80    -------81    Tensor82        Cosine Similarity Matrix83    """84    return cosine_similarity(85        mat[..., None, :, :], mat[..., :, None, :], dim=-1
...

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.