def calc_hungarian_alignment_score(claim, headline):
"""Calculate the alignment score between the two texts s and t
using the implementation of the Hungarian alignment algorithm
provided in https://pypi.python.org/pypi/munkres/."""
claim_tokens = nltk.word_tokenize(claim)
headline_tokens = nltk.word_tokenize(headline)
df = pd.DataFrame(index=claim_tokens, columns=headline_tokens, data=0.)
for c in claim_tokens:
for a in headline_tokens:
df.loc[c, a] = compute_paraphrase_score(c, a)
matrix = df.values
cost_matrix = make_cost_matrix(matrix, lambda cost: _max_ppdb_score - cost)
indices = _munk.compute(cost_matrix)
total = 0.0
for row, column in indices:
value = matrix[row][column]
total += value
# Divide total revenue by size of lower dimension (n) (no. of tokens in shorter claim/headline)
# to normalize since the algorithm will always return n pairs of indices.
return total / float(np.min(matrix.shape))
def getHungarianIndices(self, referenceChainsCount, mobileChainsCount,
matches):
"""
Run the hungarian algorithm from the Munkres module to determine
the optimal matches of chains.
Args:
referenceChainsCount: number of chains in reference
mobileChainsCount: number of chains in mobile
matches: the chain matches as determined by prody
Returns:
optimal matches based on the hungarian algorithm in the form
of indices and the corresponding matchesMatrix for the indicies
"""
profitStack = [None] * (referenceChainsCount * mobileChainsCount)
matchesStack = [None] * (referenceChainsCount * mobileChainsCount)
for element in range(0, len(matches)):
profitStack[element] = self.hungarianProfit(
matches[element][2], matches[element][3])
matchesStack[element] = matches[element]
profitMatrix = np.zeros((referenceChainsCount, mobileChainsCount))
matchesMatrix = np.zeros((referenceChainsCount, mobileChainsCount),
dtype=object)
for row in range(0, referenceChainsCount):
for column in range(0, mobileChainsCount):
profitMatrix[row][column] = profitStack.pop(0)
matchesMatrix[row][column] = matchesStack.pop(0)
profitMatrix = profitMatrix.tolist()
cost_matrix = make_cost_matrix(profitMatrix,
lambda cost: 1000.0 - cost)
m = Munkres()
indices = m.compute(cost_matrix)
return indices, matchesMatrix
def d2dAllocate(lambda_matrix, maximum_rb_allowed):
#if multiple resource blocks are allowed for d2d users then repeat the rows
if (not (maximum_rb_allowed == 1)):
lambda_matrix_new = []
for i in range(0, len(lambda_matrix)):
for j in range(0, maximum_rb_allowed):
lambda_matrix_new.append(lambda_matrix[i])
lambda_matrix = lambda_matrix_new
#convert profit matrix to cost matrix
cost_matrix = make_cost_matrix(lambda_matrix,
lambda cost: sys.maxsize - cost)
m = Munkres()
indexes = m.compute(
cost_matrix
) #indexes contains the 2d indexes of the maximum weight allocations
allocated_d2d_in_channels = np.zeros(len(lambda_matrix[0])) - 1
d2d_and_indexes = [] #indexes to return
for row, column in indexes:
allocated_d2d_in_channels[column] = int(row / maximum_rb_allowed)
allocated_d2d_in_channels = allocated_d2d_in_channels.astype(int)
for i in range(0, len(allocated_d2d_in_channels)):
if (not (allocated_d2d_in_channels[i] == -1)):
d2d_and_indexes.append([
allocated_d2d_in_channels[i],
[allocated_d2d_in_channels[i], i]
])
return d2d_and_indexes
def computePerfectMatching(self):
# transform dict to array
inputMatrix = self.transformDictToArray()
# compute perfect matching
cost_matrix = make_cost_matrix(inputMatrix,
lambda cost: sys.maxsize - cost)
m = Munkres()
perfect_indexes = m.compute(cost_matrix)
# produce the final result
# find if rowIndex is in table1 or table2
# if self.table1[0]['head']['row']:
# # index a is in table1
# for (a, b) in perfect_indexes:
# # find elements with indexes a,b in the left and right tables
#
# self.toBeJoined.append(())
# else:
# # index a is in table2
# for (a, b) in perfect_indexes
#
#
return perfect_indexes
def _find_optimal_match(self):
matrix = [[0 for _ in range(len(self._orders))]
for _ in range(len(self._couriers))]
viewed_rows = {i: 0 for i in range(len(self._couriers))}
for i, courier in enumerate(self._couriers):
for j, order in enumerate(self._orders):
revenue_from_completion = self.revenue_from_completing_order(
courier, order)
if revenue_from_completion == float('-inf'):
matrix[i][j] = DISALLOWED
else:
matrix[i][j] = revenue_from_completion
elements_to_remove = [
x for x in viewed_rows if viewed_rows[x] == len(self._orders)
]
matrix = np.delete(matrix, elements_to_remove, axis=0)
if elements_to_remove:
self._couriers = [
x for idx, x in enumerate(self._couriers)
if idx not in elements_to_remove
]
if matrix.size == 0:
return None
cost_matrix = make_cost_matrix(
matrix, lambda cost: (sys.maxsize - cost)
if (cost != DISALLOWED) else DISALLOWED)
indexes = self.m.compute(cost_matrix)
if not indexes:
return None
return indexes
def munkres_score(gt, pred):
"""
:param gt: a list of lists, each containing ints
:param pred: a list of lists, each containing ints
:return: accuracy
"""
# Combine all the sequences into one long sequence for both gt and pred
gt_combined = np.concatenate(gt)
pred_combined = np.concatenate(pred)
# Make sure we're comparing the right shapes
assert(gt_combined.shape == pred_combined.shape)
# Build out the contingency matrix
# This follows the methodology suggested by Zhou, De la Torre & Hodgkins, PAMI 2013.
mat = contingency_matrix(gt_combined, pred_combined)
# We need to make the cost matrix
# We use the fact that no entry can exceed the total length of the sequence
cost_mat = make_cost_matrix(mat, lambda x: gt_combined.shape[0] - x)
# Apply the Munkres method (also called the Hungarian method) to find the optimal cluster correspondence
m = Munkres()
indexes = m.compute(cost_mat)
# Pull out the associated 'costs' i.e. the cluster overlaps for the correspondences we've found
cluster_overlaps = mat[list(zip(*indexes))]
# Now compute the accuracy
accuracy = np.sum(cluster_overlaps)/float(np.sum(mat))
return accuracy
def calc_hungarian_alignment_score(self, s, t):
"""Calculate the alignment score between the two texts s and t
using the implementation of the Hungarian alignment algorithm
provided in https://pypi.python.org/pypi/munkres/."""
s_toks = get_tokenized_lemmas(s)
t_toks = get_tokenized_lemmas(t)
#print("#### new ppdb calculation ####")
#print(s_toks)
#print(t_toks)
df = pd.DataFrame(index=s_toks, columns=t_toks, data=0.)
for c in s_toks:
for a in t_toks:
df.ix[c, a] = self.compute_paraphrase_score(c, a)
matrix = df.values
cost_matrix = make_cost_matrix(matrix, lambda cost: _max_ppdb_score - cost)
indexes = _munk.compute(cost_matrix)
total = 0.0
for row, column in indexes:
value = matrix[row][column]
total += value
#print(s + ' || ' + t + ' :' + str(indexes) + ' - ' + str(total / float(np.min(matrix.shape))))
# original procedure returns indexes and score - i do not see any use for the indexes as a feature
# return indexes, total / float(np.min(matrix.shape))
return total / float(np.min(matrix.shape))
def score(self, seq_gt, seq_pred):
seq_gt = self.prep_seq(seq_gt)
seq_pred = self.prep_seq(seq_pred)
m, n = len(seq_gt), len(seq_pred) # length of two sequences
if m == 0:
return 1.
if n == 0:
return 0.
similarities = torch.zeros((m, n))
for i in range(m):
for j in range(n):
a = self.vectors[seq_gt[i]]
b = self.vectors[seq_pred[j]]
a = torch.from_numpy(a)
b = torch.from_numpy(b)
similarities[i, j] = torch.mean(
F.cosine_similarity(a.unsqueeze(0),
b.unsqueeze(0))).unsqueeze(-1)
similarities = (similarities + 1) / 2
similarities = similarities.numpy()
ass = self.munkres.compute(munkres.make_cost_matrix(similarities))
intersection_score = .0
for a in ass:
intersection_score += similarities[a]
iou_score = intersection_score / (m + n - intersection_score)
return iou_score
def similar(list_current,list_called): global P,total # comparing the length of two files to compare smaller length to bigger if len(list_current)<len(list_called): # calling comparison function to compare both files line by line for similarity similarity = comparison(list_current,list_called) # storing the lenght of smaller P = len(list_current) point=[[0 for x in range(len(list_called))] for y in range(len(list_current))] else: # calling comparison function to compare both files line by line for similarity similarity = comparison(list_called,list_current) P = len(list_called) point=[[0 for x in range(len(list_current))] for y in range(len(list_called))] # calling functions of munkres to form maximum weighted bipartite matching graph graph_matrix = make_cost_matrix(similarity, lambda cost: 1.0 - cost) m = Munkres() indexes =m.compute(graph_matrix) total = 0 for row, column in indexes: # forming list of points(lines) of similarity between two files value = similarity[row][column] if value>0.0: total += 1 point[row][column]=1 return point
def make_cost_matrix(profit_matrix, inversion_function):
"""
**DEPRECATED**
Please use the module function ``make_cost_matrix()``.
"""
import munkres
return munkres.make_cost_matrix(profit_matrix, inversion_function)
def maximum_weight_bipartite(matrix):
cost_matrix = make_cost_matrix(matrix, lambda cost: 100000 - cost)
m = Munkres()
indices = m.compute(cost_matrix)
return indices
def maxi():
#matrix = [[20, 16, 22, 18],
# [25, 28, 15, 21],
# [27, 20, 23, 26],
# [24, 22, 23, 22]]
f = open('newds.txt')
n = int(f.readline())
matrix = []
for i in range(n):
list1 = map(int, (f.readline()).split())
matrix.append(list1)
cost_matrix = make_cost_matrix(matrix)
m = Munkres()
indexes = m.compute(cost_matrix)
print_matrix(matrix, msg='Highest profits through this matrix:')
total = 0
for row, column in indexes:
value = matrix[row][column]
total += value
print('(%d, %d) -> %d' % (row, column, value))
print('total profit=%d' % total)
start_time = time.clock()
timex = time.clock() - start_time
time2 = ("Time =", timex, "seconds")
print(time2)
#timex1 = string(timex)
timetext = open('times2.txt', 'a+')
timetext.write(str(timex))
main()
def filter_rects(all_rects,
threshold,
net_config,
input_rects=[],
max_threshold=1.0):
"""Takes in all_rects and based on the threshold carries out the stitching process
as described in the paper."""
accepted_rects = input_rects
for i in range(0, net_config["grid_height"], 1):
for j in range(0, net_config["grid_width"], 1):
relevant_rects = []
current_rects = [
r for r in all_rects[i][j] if r.confidence > threshold
]
for other in accepted_rects:
for current in current_rects:
if other.overlaps(current):
relevant_rects.append(other)
break
if len(relevant_rects) == 0 or len(current_rects) == 0:
accepted_rects += current_rects
continue
matrix = []
for c in current_rects:
row = []
for a in relevant_rects:
row.append(1000)
if a.overlaps(c):
row[-1] -= 100
row[-1] += a.distance(c) / 1000.0
matrix.append(row)
m = Munkres()
cost_matrix = make_cost_matrix(matrix, lambda x: x)
indices = m.compute(matrix)
bad = set()
for row, column in indices:
c = current_rects[row]
a = relevant_rects[column]
if c.confidence > max_threshold:
bad.add(row)
continue
if c.overlaps(a):
if c.confidence > a.confidence and c.iou(a) > 0.7:
c.true_confidence = a.confidence
accepted_rects.remove(a)
else:
bad.add(row)
for k in range(len(current_rects)):
if k not in bad:
accepted_rects.append(current_rects[k])
return accepted_rects
def KM_mapping(action, VEHICLES, request_selected, vehicle,
current_time): # ?????
'''
:param prob_weights: a_prob
:return: matching final action,"indexes":
'''
print('cal_profit begin')
profit_matrix = cal_profit(VEHICLES, request_selected, vehicle,
current_time)
print('cal_profit end')
action = action.reshape([1, VEHICLES_NUMS * REQUEST_NUMS])
action = np.apply_along_axis(lambda x: round(x[0], 2), 0, action)
action_weights = action.reshape([VEHICLES_NUMS, REQUEST_NUMS])
km_matrix = profit_matrix * action_weights
# km_weights = make_cost_matrix(km_matrix, lambda item: (maxsize - item) if item != 0 else DISALLOWED)
km_weights = make_cost_matrix(km_matrix, lambda item: (400 - item))
m = Munkres()
print('km begin')
indexes = m.compute(km_weights)
print('km_end')
print_matrix(profit_matrix, msg='Highers profit through this matrix:')
total = 0
temp_indexes = copy.deepcopy(indexes)
for row, column in temp_indexes:
value = profit_matrix[row][column]
if value == 0:
indexes.remove((row, column))
total += value
return indexes, total
# KM_mapping([])
def KM_mapping(action, REQUESTS, VEHICLES, request_selected, vehicle,
current_time): # ?????
'''
:param prob_weights: a_prob
:return: matching final action,"indexes":
'''
# prob_weights = [5, DISALLOWED, 70, 0,
# 10, 3, 2, 3,
# 9, DISALLOWED, 4, 5,
# 1,2,3,4,
# 90,5,1,DISALLOWED]
profit_matrix = cal_profit(REQUESTS, VEHICLES, request_selected, vehicle,
current_time)
km_matrix = profit_matrix * action.reshape([VEHICLES_NUMS, REQUEST_NUMS])
km_weights = make_cost_matrix(
km_matrix, lambda item: (maxsize - item) if item != 0 else DISALLOWED)
# matrix = np.array(prob_weights)
# matrix = np.reshape(matrix, [VEHICLES_NUMS, REQUEST_NUMS])
# matrix = matrix.transpose()
# cost_matrix = make_cost_matrix(matrix, lambda cost: (maxsize - cost) if (cost != DISALLOWED) else DISALLOWED)
m = Munkres()
indexes = m.compute(km_weights)
print_matrix(profit_matrix, msg='Highers profit through this matrix:')
total = 0
for row, column in indexes:
# print(row, column)
value = profit_matrix[row][column]
total += value
# print('(%d, %d) -> %d' % (row, column, value))
# print('total profit: %d' % total)
return indexes, total
# KM_mapping([])
def get_suitability_score(customer_list, product_list):
'''
calculate the total maximum suitability score
by using munkres algorithm and returns a detailed
customer_product_enties & total suitability score
'''
suitability_scores = []
customer_suitability_scores = []
for customer in customer_list:
for product in product_list:
customer_suitability_scores.append(SuitabilityScore.calculate_suitability_score(customer,product))
suitability_scores.append(customer_suitability_scores)
customer_suitability_scores = []
customer_product_entries = []
cost_matrix = make_cost_matrix(suitability_scores, lambda cost: 1e10 - cost)
munkres = Munkres()
indexes = munkres.compute(cost_matrix)
total_suitability_score = 0
for customer_index, product_index in indexes:
suitability_score = suitability_scores[customer_index][product_index]
total_suitability_score += suitability_score
suitability_score_entry = SuitabilityScoreEntry(customer_list[customer_index],product_list[product_index],suitability_score)
customer_product_entries.append(suitability_score_entry)
#print(customer_index,product_index)
return customer_product_entries,total_suitability_score
def make_cost_matrix(profit_matrix, inversion_function):
"""
**DEPRECATED**
Please use the module function ``make_cost_matrix()``.
"""
import munkres
return munkres.make_cost_matrix(profit_matrix, inversion_function)
def ngramset_edit_distance(set1, set2):
def get_yxgraph_distance(x, y):
import math
if (x == y):
return 0
elif (x > y):
return math.sqrt(math.pow((x - y), 2))
else:
return -math.sqrt(math.pow((y - x), 2))
matrix = _ngram_matrix(set1, set2)
# with open("matrix", 'wb') as file:
# file.write(str(matrix))
cost_matrix = make_cost_matrix(matrix, lambda cost: sys.maxint - cost)
m = Munkres()
indexes = m.compute(cost_matrix)
# total = 0.0
max_matrix = []
xygraph_distance_list = []
for row, column in indexes:
value = matrix[row][column]
max_matrix.append(value)
xygraph_distance_list.append(get_yxgraph_distance(row, column))
# total += value
edit_distance = numpy.mean(max_matrix) / 100
variance = numpy.var(max_matrix)
# if edit_distance > 0.7:
# sim2 = _similarity(xygraph_distance_list, 2)
# sim3 = _similarity(xygraph_distance_list, 3)
# return edit_distance, variance, sim2, sim3, xygraph_distance_list
return edit_distance, variance, None, None, xygraph_distance_list
def perform_alignment(ref_instances, sys_instances, kernel, maximize=True):
disallowed = {}
max_sim = 0
sim_matrix, component_matrix = [], []
for s_i, s in enumerate(sys_instances):
sim_row = []
comp_row = []
for r_i, r in enumerate(ref_instances):
sim, comp = kernel(r, s)
sim_row.append(sim)
comp_row.append(comp)
if sim == DISALLOWED:
disallowed[(s_i, r_i)] = True
else:
if sim > max_sim: max_sim = sim
sim_matrix.append(sim_row)
component_matrix.append(comp_row)
if maximize:
def _mapper(sim):
return max_sim + 1 if sim == DISALLOWED else (max_sim + 1) - sim
else:
def _mapper(sim):
return max_sim + 1 if sim == DISALLOWED else sim
matrix = make_cost_matrix(sim_matrix, _mapper)
correct_detects, false_alarms, missed_detects = [], [], []
unmapped_sys = set(range(0, len(sys_instances)))
unmapped_ref = set(range(0, len(ref_instances)))
if len(matrix) > 0:
for s_i, r_i in Munkres().compute(matrix):
if disallowed.get((s_i, r_i), False):
continue
unmapped_sys.remove(s_i)
unmapped_ref.remove(r_i)
correct_detects.append(
AlignmentRecord(ref_instances[r_i], sys_instances[s_i],
sim_matrix[s_i][r_i],
component_matrix[s_i][r_i]))
for r_i in unmapped_ref:
missed_detects.append(
AlignmentRecord(ref_instances[r_i], None, None, None))
for s_i in unmapped_sys:
false_alarms.append(
AlignmentRecord(None, sys_instances[s_i], None, None))
return (correct_detects, missed_detects, false_alarms)
def test_profit():
profit_matrix = [[94, 66, 100, 18, 48], [51, 63, 97, 79, 11],
[37, 53, 57, 78, 28], [59, 43, 97, 88, 48],
[52, 19, 89, 60, 60]]
import sys
cost_matrix = munkres.make_cost_matrix(profit_matrix,
lambda cost: sys.maxsize - cost)
indices = m.compute(cost_matrix)
profit = sum([profit_matrix[row][column] for row, column in indices])
assert_equals(profit, 392)
def max_match(matrix): cost_matrix = make_cost_matrix(matrix, lambda cost: 100.0-cost) m = Munkres() indexes = m.compute(cost_matrix) #print_matrix(matrix, msg='Lowest cost through this matrix:') total = 0 for row, column in indexes: value = matrix[row][column] total += value #print '(%d, %d) -> %d' % (row, column, value) return total
def filter_rects(all_rects, threshold, input_rects=[], max_threshold=1.0, config=None):
"""Takes in all_rects and based on the threshold carries out the stitching process
as described in the paper."""
accepted_rects = input_rects
for i in range(0, config["grid_height"], 1):
for j in range(0, config["grid_width"], 1):
relevant_rects = []
current_rects = [r for r in all_rects[i][j] if r.confidence > threshold]
for other in accepted_rects:
for current in current_rects:
if other.overlaps(current):
relevant_rects.append(other)
break
if len(relevant_rects) == 0 or len(current_rects) == 0:
accepted_rects += current_rects
continue
matrix = []
for c in current_rects:
row = []
for a in relevant_rects:
row.append(1000)
if a.overlaps(c):
row[-1] -=100
row[-1] += a.distance(c) / 1000.0
matrix.append(row)
m = Munkres()
cost_matrix = make_cost_matrix(matrix, lambda x: x)
indices = m.compute(matrix)
bad = set()
for row, column in indices:
c = current_rects[row]
a = relevant_rects[column]
if c.confidence > max_threshold:
bad.add(row)
continue
if c.overlaps(a):
if c.confidence > a.confidence and c.iou(a) > 0.7:
c.true_confidence = a.confidence
accepted_rects.remove(a)
else:
bad.add(row)
for k in range(len(current_rects)):
if k not in bad:
accepted_rects.append(current_rects[k])
return accepted_rects
def compute_matches(affinity_scores, j1_pts, j2_pts):
matching_results = []
match_confidence_threshold = CONFIG.match_confidence_threshold
j1_count, j2_count = affinity_scores.shape
indices = MUNKRES_INSTANCE.compute(make_cost_matrix(affinity_scores.tolist(), inversion_function=lambda x : 2 - x))
for row,col in indices:
if(affinity_scores[row,col]>match_confidence_threshold):
matching_results.append((j1_pts[row], j2_pts[col], affinity_scores[row,col]))
return matching_results
def generate_transformed_matrix(self):
confusion = self.mat
confusion = confusion.T
cost_matrix = munkres.make_cost_matrix(
confusion, lambda cost: sys.long_info.sizeof_digit - cost)
m = munkres.Munkres()
indexes = m.compute(cost_matrix)
new_mat = np.zeros(confusion.shape)
for i in range(len(indexes)):
new_mat[:, i] = confusion[:, indexes[i][1]]
return new_mat
def return_munkres_result(matrix):
cost_matrix = make_cost_matrix(matrix, lambda cost: (sys.maxsize - cost) if (cost != DISALLOWED) else DISALLOWED)
m = Munkres()
indexes = m.compute(cost_matrix)
bestMatches = []
for row, column in indexes:
value = matrix[row][column]
bestMatches.append(value)
return bestMatches
def show_csv_file(request, csv_id):
data = CsvFile.objects.filter(id=csv_id).first().text.strip()
mentors = set()
mentees = set()
lines = data.splitlines()
for line in lines[1:]:
if len(line) <= 1:
continue
print(line)
mentor, mentee, utility = [x.strip() for x in line.split(',')]
mentors.add(mentor)
mentees.add(mentee)
matrix = [[DISALLOWED] * len(mentees) for x in mentors]
mentors = {x[1]: x[0] for x in enumerate(sorted(mentors))}
mentees = {x[1]: x[0] for x in enumerate(sorted(mentees))}
for line in lines[1:]:
if len(line) <= 1:
continue
mentor, mentee, utility = [x.strip() for x in line.split(',')]
utility = float(utility)
matrix[mentors[mentor]][mentees[mentee]] = utility
cost_matrix = make_cost_matrix(
matrix, lambda cost: (200 - cost)
if (cost != DISALLOWED) else DISALLOWED)
m = Munkres()
indexes = m.compute(cost_matrix)
mentors = {v: k for k, v in mentors.items()}
mentees = {v: k for k, v in mentees.items()}
used_mentors = set()
used_mentees = set()
table = []
for row, column in indexes:
value = matrix[row][column]
mentor = mentors[row]
mentee = mentees[column]
table.append([mentor, mentee, value])
used_mentors.add(mentor)
used_mentees.add(mentee)
mentors = set([x for x in mentors.values()])
mentees = set([x for x in mentees.values()])
unused_mentees = mentees - used_mentees
unused_mentors = mentors - used_mentors
table = sorted(table, key=lambda x: x[0])
context = {
'table': table,
'unused_mentors': unused_mentors,
'unused_mentees': unused_mentees
}
return render(request, 'show_table.html', context=context)
def compute_agreements(similarity):
import munkres
import numpy as np
m = munkres.Munkres()
print("Computing mapping...")
similarity = munkres.make_cost_matrix(similarity, lambda cost: 1 - cost)
indexes = m.compute(similarity)
agreement = np.sum([1 - similarity[r][c] for r, c in indexes]) / len(similarity)
print("Agreement:", agreement)
return agreement
def test_profit_float():
profit_matrix = [[94.01, 66.02, 100.03, 18.04, 48.05],
[51.06, 63.07, 97.08, 79.09, 11.1],
[37.11, 53.12, 57.13, 78.14, 28.15],
[59.16, 43.17, 97.18, 88.19, 48.2],
[52.21, 19.22, 89.23, 60.24, 60.25]]
import sys
cost_matrix = munkres.make_cost_matrix(profit_matrix,
lambda cost: sys.maxsize - cost)
indices = m.compute(cost_matrix)
profit = sum([profit_matrix[row][column] for row, column in indices])
assert profit == pytest.approx(362.65)
def calculateSimilarity(sent1, sent2):
# Case Correction
sent1 = sent1.lower()
sent2 = sent2.lower()
# Tokenization
tokens1 = word_tokenize(sent1)
tokens2 = word_tokenize(sent2)
# Remove punctuations
tokens1 = [x for x in tokens1 if x not in string.punctuation]
tokens2 = [x for x in tokens2 if x not in string.punctuation]
# Remove Stopwords
# stopWords = set(stopwords.words('english'))
# tokens1 = [x for x in tokens1 if x not in stopWords]
# tokens2 = [x for x in tokens2 if x not in stopWords]
# Lemmatization
# lemmatizer = WordNetLemmatizer()
# tokens1 = [lemmatizer.lemmatize(x) for x in tokens1]
# tokens2 = [lemmatizer.lemmatize(x) for x in tokens2]
# Model has token?
tokens1 = [x for x in tokens1 if x in model.vocab]
tokens2 = [x for x in tokens2 if x in model.vocab]
if len(tokens1) > 0 and len(tokens2) > 0:
m = Munkres()
pairMatrix = []
for t1 in tokens1:
tmpList = []
for t2 in tokens2:
tmpList.append(100 * JSD(model[t1], model[t2]))
pairMatrix.append(tmpList)
cost_matrix = make_cost_matrix(pairMatrix, lambda cost: 100 - cost)
indexes = m.compute(cost_matrix)
# print_matrix(pairMatrix, msg='Lowest cost through this matrix:')
total = 0
for row, column in indexes:
value = pairMatrix[row][column]
total += value
# print('(%d, %d) -> %d' % (row, column, value))
# print('total cost: %d' % total)
# print(total / len(indexes))
return total / len(indexes) / 100
# return 2 * total / (len(tokens1) + len(tokens2)) / 100
else:
return 0
def KM_mapping(REQUESTS, VEHICLES, request_selected, vehicle, current_time):
profit_matrix = cal_profit(REQUESTS, VEHICLES, request_selected, vehicle,
current_time)
km_weights = make_cost_matrix(
profit_matrix, lambda item: (maxsize - item)
if item != 0 else DISALLOWED)
m = Munkres()
indexes = m.compute(km_weights)
total = 0
for row, column in indexes:
value = profit_matrix[row][column]
total += value
return indexes, total
def compute_agreements(similarity):
import munkres
import numpy as np
m = munkres.Munkres()
print("Computing mapping...")
similarity = munkres.make_cost_matrix(similarity, lambda cost: 1 - cost)
indexes = m.compute(similarity)
agreement = np.sum([1 - similarity[r][c]
for r, c in indexes]) / len(similarity)
print("Agreement:", agreement)
return agreement
def test_profit():
profit_matrix = [[94, 66, 100, 18, 48],
[51, 63, 97, 79, 11],
[37, 53, 57, 78, 28],
[59, 43, 97, 88, 48],
[52, 19, 89, 60, 60]]
import sys
cost_matrix = munkres.make_cost_matrix(
profit_matrix, lambda cost: sys.maxsize - cost
)
indices = m.compute(cost_matrix)
profit = sum([profit_matrix[row][column] for row, column in indices])
assert_equals(profit, 392)
def calculateIdealComposition(lineup, composition):
m = Munkres()
matrix = buildMatrix(lineup, composition)
cost_matrix = munkres.make_cost_matrix(matrix, lambda cost: int(100 - (cost * 100)))
indexes = m.compute(cost_matrix)
picks = []
for index in indexes:
picks.append((lineup[index[0]], composition[index[1]]))
return picks
def cluster_assignment(id2label_1, id2label_2):
"""
Assignes cluster names from first clusterisation to cluster names of second clusterisation
using Hungarian algorithm of the assignment problem (maximising the similarity between clusters
from different clusterisations)
Labels could have different lengths and contain different types of objects
Arguments:
id2label_1: dict of id to label from first clusterisation
id2label_2: dict of id to label from second clusterisation
Returns:
dict: how to transform first labels to second
"""
# Get intersection of two dicts
ids = []
labels_1, labels_2 = [], []
for news_id, label_1 in id2label_1.items():
if news_id in id2label_2:
ids.append(news_id)
labels_1.append(label_1)
labels_2.append(id2label_2[news_id])
n_clusters_1 = len(set(labels_1))
n_clusters_2 = len(set(labels_2))
# Encode labels so they would be from 0 to max
l_encoder_1 = LabelEncoder().fit(labels_1)
labels_1_transf = l_encoder_1.transform(labels_1)
l_encoder_2 = LabelEncoder().fit(labels_2)
labels_2_transf = l_encoder_2.transform(labels_2)
# Create matrix of distances between two clusters
matrix = [[0] * n_clusters_2 for i in range(n_clusters_1)]
for i in range(min(len(labels_1), len(labels_2))):
matrix[labels_1_transf[i]][labels_2_transf[i]] += 1
# Compute Munkres (Hungarian) algorithm
cost_matrix = make_cost_matrix(matrix, lambda cost: sys.maxsize - cost)
m = Munkres()
indices = m.compute(cost_matrix)
# Transform labels back
transform = {}
for (label_1, label_2) in indices:
transform[l_encoder_1.inverse_transform(
label_1)] = l_encoder_2.inverse_transform(label_2)
return transform
def _find_best_permutation(self, spectral, spatial, idx_constant):
'''Finds the best permutation of classes in spectral and spatial model.
Args:
spectral (np.array): Conditional log-likelihood of spectral model
(as in Eq.19) in [1], shape (N,T,F)
spatial (np.array): Conditional log-likelihood of spatial model
(as in Eq.19) in [1], shape (N,T,F)
idx_constant (tuple or int) indices of axis which have constant permutation
Examples:
idx_constant = (1,2)
-> for all time frames and frequency bins
the permutation is constant
(finding 1 global permutation)
idx_constant = 1
-> for all time frames the permutation is constant
(finding permutation for each frequency)
Returns:
permutations (dict): mapping tuples of time and frequency indices
to the best permutation. For constant indices,
the map contains only index 0.
Examples:
permutations = {(0,0) : [2, 0, 1]}
-> one global permutation (idx_constant = (1,2))
-> spectral comp. 0 corresponds to spatial comp. 2
-> spectral comp. 1 corresponds to spatial comp. 0
-> spectral comp. 2 corresponds to spatial comp. 1
[1] Integration of variational autoencoder and spatial clustering
for adaptive multi-channel neural speech separation;
K. Zmolikova, M. Delcroix, L. Burget, T. Nakatani, J. Cernocky
'''
if isinstance(idx_constant, int):
idx_constant = (idx_constant, )
idx_constant = tuple([i + 1 for i in idx_constant])
perm_scores = logsumexp(spectral[:, None, :, :] +
spatial[None, :, :, :],
axis=idx_constant)
perm_scores = np.expand_dims(perm_scores, idx_constant)
permutations = {}
for i1, i2 in np.ndindex(perm_scores.shape[-2:]):
idx_perm = Munkres().compute(
make_cost_matrix(perm_scores[:, :, i1, i2]))
idx_perm.sort(key=lambda x: x[0])
permutations[i1, i2] = [i[1] for i in idx_perm]
return permutations
def matching(male_distances):
cost_matrix = munkres.make_cost_matrix(male_distances,
lambda cost: 1.0 - cost)
m = Munkres()
indices = m.compute(cost_matrix)
total = 0.0
pairings = {}
for row, column in indices:
value = male_distances[row][column]
total += value
pairings[column] = [row]
#print "Satish was here"
print 'total profit=%f' % total
return pairings
def symmetric_matching(table):
"""Cluster matching that maximizes the total number of elements in common
while matching at most one cluster to another.
>>> symmetric_matching([[2, 2, 1], [1, 0, 2]])
[(0, 0), (1, 2)]
>>> symmetric_matching([[3, 2, 1], [2, 0, 1]])
[(0, 0), (1, 2)]
"""
maximum = max(it.chain(*table))
cost_matrix = make_cost_matrix(table, lambda x: maximum - x)
indices = Munkres().compute(cost_matrix)
return indices
def make_match(advertisers, persons, ctrfunc = funky_ctr):
# make the ctr matrix
ctr_matrix = make_ctr_matrix(advertisers, persons, ctrfunc)
# convert it to a cost matrix by subtracting all values from a larger value
cost_matrix = make_cost_matrix(ctr_matrix, lambda ctr: sys.maxsize - ctr)
# compute the match
match = Munkres().compute(cost_matrix)
# elements in match are two-element lists, where the first is the
# index into advertisers and second is index into persons
# compute the total ctr by looking up the match elements in the ctr
# matrix
total_ctr = sum(map(lambda pair: ctr_matrix[pair[0]][pair[1]], match))
return match, total_ctr
def do_hungarian_assignment(dict_a, dict_b, cost_func, yield_condt, cost_matrix_func=None):
for ka in dict_a.keys():
if ka in dict_b:
ia = dict_a[ka]
ib = dict_b[ka]
mat = [ [cost_func(a,b) for b in ib] for a in ia]
#max similarity calculation
c_mat = None
if cost_matrix_func is not None:
c_mat = make_cost_matrix(mat,cost_matrix_func)
else:
c_mat = mat
indexes = MUNKR.compute(c_mat)
for row, col in indexes:
# yield only if condition satisfied
if yield_condt(mat[row][col]):
#print '(%d, %d) -> %d' % (row, col, mat[row][col])
yield ia[row], ib[col]
def assignDuties(dayA, dayB):
if len(dayA) != len(dayB):
print "Illegal!! number of duties should be equal every day"
exit(1)
matrix = createMatrix(len(dayA))
##Fill the matrix with the similarities
for i in range(len(dayA)):
for j in range(len(dayA)):
matrix[i][j] = calcSimilarity(dayA[i], dayB[j])
##The following line is called to make sure we find the maximum sum and not the minimum
##Note that since all similarities are in [0,1] 2 is bigger than all.
cost_matrix = make_cost_matrix(matrix, lambda cost: 2 - cost)
m = Munkres()
##Indexes will contain the assignment
indexes = m.compute(cost_matrix)
res = []
for row, column in indexes:
res.append(("Driver "+str(row + 1) ,("Day 1 duty: "+str(row + 1),"Day 2 duty: "+str(column + 1))))
return res
def get_best_matching(source_corpus, target_corpus, scores):
stripper = LanguageStripper()
err = 0
m = munkres.Munkres()
cost_matrix = munkres.make_cost_matrix(scores, lambda cost: 1 - cost)
indexes = m.compute(cost_matrix)
for row, column in indexes:
s_url = source_corpus.keys()[row]
t_url = target_corpus.keys()[column]
success = stripper.strip(t_url) == stripper.strip(s_url)
if not success:
err += 1
# sys.stdout.write("%f\t%s\t%s\t%s\n" %
# (scores[row, column], success, s_url, t_url))
n = min(len(source_corpus), len(target_corpus))
sys.stderr.write("Correct: %d out of %d = %f%%\n" %
(n - err, n, (1. * n - err) / n))
def hungarianAssignment(cases, controls, numberOfControlsPerCase):
selectedControls = list()
convFactor = 100000.00
m = list()
for control in controls:
row = list()
for _ in xrange(numberOfControlsPerCase):
row += [(control.relatedTo.get(case) if (control.relatedTo.has_key(case)) else 0) for case in cases]
m.append(row)
cm = munkres.make_cost_matrix(m, lambda cost: convFactor - cost * convFactor)
matrix = cm
m = Munkres()
indexes = m.compute(matrix)
total = 0
for row, column in indexes:
value = matrix[row][column]
total += value
if value < convFactor:
selectedControls.append(controls[row % len(controls)])
print("\tassignment kic score: %f" % (float(len(indexes) * convFactor - total) / convFactor))
print("\tall kic score: %f" % (kicScore(cases, selectedControls)))
return [person.id for person in selectedControls]
def calc_hungarian_alignment_score(s, t):
"""Calculate the alignment score between the two texts s and t
using the implementation of the Hungarian alignment algorithm
provided in https://pypi.python.org/pypi/munkres/."""
s_toks = get_tokenized_lemmas(s)
t_toks = get_tokenized_lemmas(t)
df = pd.DataFrame(index=s_toks, columns=t_toks, data=0.)
for c in s_toks:
for a in t_toks:
df.ix[c, a] = compute_paraphrase_score(c, a)
matrix = df.values
cost_matrix = make_cost_matrix(matrix, lambda cost: _max_ppdb_score - cost)
indexes = _munk.compute(cost_matrix)
total = 0.0
for row, column in indexes:
value = matrix[row][column]
total += value
return indexes, total / float(np.min(matrix.shape))
def linkTemplates(self, sentence):
""" link group size and group templates using Hungarian matching algorithm """
templates = sentence.templates
qTemplateList = templates.getList(self.quantityType)
mTemplateList = templates.getList(self.mentionType)
nQuantities = len(qTemplateList)
nMentions = len(mTemplateList)
maxSize = max(nQuantities, nMentions)
if nQuantities == 0 or nMentions == 0:
return
probMatrix = []
for qIdx in range(maxSize):
probMatrix.append([])
for mIdx in range(maxSize):
probMatrix[qIdx].append(0)
for fv in templates.featureVectors:
probMatrix[fv.valueId][fv.mentionId] = fv.prob * 1000
costMatrix = munkres.make_cost_matrix(probMatrix, lambda cost: 1000 - cost)
m = munkres.Munkres()
# print probMatrix
# print costMatrix
indices = m.compute(costMatrix)
for qIdx, mIdx in indices:
if qIdx < nQuantities and mIdx < nMentions:
prob = probMatrix[qIdx][mIdx]
if prob >= 500:
# this quantity and mention should be associated
prob = float(prob) / 1000
qTemplate = qTemplateList[qIdx]
mTemplate = mTemplateList[mIdx]
self.linkQuantityAndMention(qTemplate, mTemplate, prob)
lda=np.reshape(lda,[145*145])
##Adding dummy label from 0-16
gmm_vote=np.append(gmm,[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16])
lda_vote=np.append(lda,[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16])
cmeans_vote=np.append(cmeans,[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16])
kmeans_vote=np.append(kmeans,[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16])
isodata_vote=np.append(isodata,[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16])
svm_vote=np.append(svm,[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16])
###Construct Label Map with K-Means as the baseline
#Construct Label Map between K-Means and LDA
matrix=contingency_matrix(kmeans_vote,lda_vote)
matrix = matrix.tolist()
cost_matrix = make_cost_matrix(matrix, lambda cost: sys.maxsize - cost)
m = Munkres()
indexes = m.compute(cost_matrix)
total = 0
newlabelA=[]
for row, column in indexes:
value = matrix[row][column]
total += value
newlabelA.append(column)
#Construct Label Map between K-Means and Gausian Mixture Model
matrix=contingency_matrix(kmeans_vote,gmm_vote)
matrix = matrix.tolist()
cost_matrix = make_cost_matrix(matrix, lambda cost: sys.maxsize - cost)
m = Munkres()
indexes = m.compute(cost_matrix)
matrix3 = ([[7, 53, 183, 439, 863, 497, 383, 563, 79, 973, 287, 63, 343, 169, 583],
[627, 343, 773, 959, 943, 767, 473, 103, 699, 303, 957, 703, 583, 639, 913],
[447, 283, 463, 29, 23, 487, 463, 993, 119, 883, 327, 493, 423, 159, 743],
[217, 623, 3, 399, 853, 407, 103, 983, 89, 463, 290, 516, 212, 462, 350],
[960, 376, 682, 962, 300, 780, 486, 502, 912, 800, 250, 346, 172, 812, 350],
[870, 456, 192, 162, 593, 473, 915, 45, 989, 873, 823, 965, 425,329, 803],
[973, 965, 905, 919, 133, 673, 665, 235, 509, 613, 673, 815, 165, 992, 326],
[322, 148, 972, 962, 286, 255, 941, 541, 265, 323, 925, 281, 601, 95, 973],
[445, 721, 11, 525, 473, 65, 511, 164, 138, 672, 18, 428, 154, 448, 848],
[414, 456, 310, 312, 798, 104, 566, 520, 302, 248, 694, 976, 430, 392, 198],
[184, 829, 373, 181, 631, 101, 969, 613, 840, 740, 778, 458, 284, 760, 390],
[821, 461, 843, 513, 17, 901, 711, 993, 293, 157, 274, 94, 192, 156, 574],
[34, 124, 4, 878, 450, 476, 712, 914, 838, 669, 875, 299, 823, 329, 699],
[815, 559, 813, 459, 522, 788, 168, 586, 966, 232, 308, 833, 251, 631, 107],
[813, 883, 451, 509, 615, 77, 281, 613, 459, 205, 380, 274, 302, 35, 805]])
costmat = make_cost_matrix(matrix3, lambda cost: sys.maxint - cost)
m = Munkres()
indexes = m.compute(costmat)
tot = 0
for row,col in indexes:
tp = matrix3[row][col]
tot += tp
print indexes
print tot
def linkTemplates(self, sentence):
""" link group size and group templates using Hungarian matching algorithm """
# print 'linking all templates'
templates = sentence.templates
onList = templates.getList('on')
erList = templates.getList('eventrate')
outcomeMeasurements = templates.getOutcomeMeasurementList()
groupList = sentence.abstract.entities.lists['group']
outcomeList = sentence.abstract.entities.lists['outcome']
nON = len(onList)
nER = len(erList)
nGroups = len(groupList)
nOutcomes = len(outcomeList)
nGroupOutcomePairs = nGroups * nOutcomes
if (nON + nER) == 0 or nGroupOutcomePairs == 0:
return # missing key information cannot make any associations
goPairs = []
for group in groupList:
for outcome in outcomeList:
# goPairs.append((group,outcome))
(nMatched, nUnmatched1, nUnmatched2) = group.partialSetMatch(outcome)
if self.groupOutcomeOverlap(group, outcome) == False:
# overlap between group/outcome may be no more than ONE word and this may be no more than 1/3 of smaller mention
goPairs.append((group,outcome))
else:
print sentence.abstract.id, '#### skipping:', group.rootMention().name, ';', outcome.rootMention().name
nGroupOutcomePairs = len(goPairs)
if nGroupOutcomePairs == 0:
return # missing key information cannot make any associations
# get unmatched event rates and outcome numbers
unmatchedON = []
unmatchedER = []
for om in outcomeMeasurements:
on = om.getOutcomeNumber()
er = om.getTextEventRate()
if er != None and on == None:
# unmatched event rate
unmatchedER.append(er)
elif on != None and er == None:
# unmatched number of outcomes
unmatchedON.append(on)
# identify as of yet unmatched event rates and outcome numbers that could potentially match each other
erMatches = {}
onMatches = {}
for on in unmatchedON:
onMatches[on] = []
for er in unmatchedER:
erMatches[er] = []
for on in unmatchedON:
couldCalculateER = False
if on.hasAssociatedGroupSize():
calculatedER = on.eventRate()
couldCalculateER = True
for er in unmatchedER:
if er.equivalentEventRates(calculatedER):
onMatches[on].append(er)
erMatches[er].append(on)
else:
for group in groupList:
groupFV = on.getMatchFeatures(group)
if groupFV != None and groupFV.prob > 0:
# it is possible to associate with this group
gs = group.getSize(sentenceIndex=sentence.index)
if gs > 0:
calculatedER = on.eventRate(groupSize=gs)
couldCalculateER = True
for er in unmatchedER:
if er.equivalentEventRates(calculatedER):
onMatches[on].append(er)
erMatches[er].append(on)
if couldCalculateER == False:
outcomeMeasurements.remove(on.outcomeMeasurement)
# discard any outcome numbers that potentially match multiple event rates
for on in onMatches.keys():
if len(onMatches[on]) == 1 and len(erMatches[onMatches[on][0]]) == 1:
# this outcome number is a potential match for only one event rate
# similarly, the event rate is only a match for this outcome number
# assume they belong to same outcome measurement
erOM = er.outcomeMeasurement
on.outcomeMeasurement.addEventRate(er)
outcomeMeasurements.remove(erOM)
# now consider all possible valid ON,ER pairings
# for on in unmatchedON:
# for er in unmatchedER:
# if on.hasAssociatedGroupSize() == False or on.equivalentEventRates(er.eventRate()) == True:
# om = OutcomeMeasurement(on)
# om.addEventRate(er)
# outcomeMeasurements.append(om)
nOutcomeMeasurements = len(outcomeMeasurements)
maxSize = max(nOutcomeMeasurements, nGroupOutcomePairs)
# initialize cost matrix for matching outcome measurements with group,outcome pairs
probMatrix = []
probMultiplier = 100000
for omIdx in range(maxSize):
probMatrix.append([])
for goIdx in range(maxSize):
if omIdx < nOutcomeMeasurements and goIdx < nGroupOutcomePairs:
om = outcomeMeasurements[omIdx]
(group, outcome) = goPairs[goIdx]
er = om.getTextEventRate()
on = om.getOutcomeNumber()
if er != None:
outcomeFV = er.getMatchFeatures(outcome)
groupFV = er.getMatchFeatures(group)
if outcomeFV == None or groupFV == None:
# this quantity has no chance of being associated with either the group or outcome mention
# this can happen if all mentions for the entity appear in a sentence after the quantity
probG_ER = 0
probO_ER = 0
else:
probO_ER = outcomeFV.prob
probG_ER = groupFV.prob
else:
probG_ER = 1
probO_ER = 1
if on != None:
# this outcome measurement has an outcome number
# is this number useful? Can we compute an event rate for this group?
# If not, discard this measurement (set probability to zero).
# if so, is the event rate compatible with the textual event rate?
# If not, discard.
calculatedER = -1
gs = group.getSize(sentenceIndex=sentence.index)
outcomeFV = on.getMatchFeatures(outcome)
groupFV = on.getMatchFeatures(group)
if outcomeFV == None or groupFV == None:
# this quantity has no chance of being associated with either the group or outcome mention
probG_ON = 0
probO_ON = 0
else:
probO_ON = outcomeFV.prob
probG_ON = groupFV.prob
if on.hasAssociatedGroupSize() == False:
# there is no group size already associated with the outcome number
# does the group have a group size?
# If so, is the resulting event rate compatible with the text one?
if gs <= 0 and er == None:
# there is no way to compute an event rate with this outcome measurement.
# it does not add any useful information.
# discard it by setting probability to zero
probG_ON = 0
probO_ON = 0
elif gs > 0:
# the proposed group has an associated size
# we can compute an event rate for this group/outcome
calculatedER = on.eventRate(groupSize=gs)
if (er != None and er.equivalentEventRates(calculatedER) == False) or abs(calculatedER) > 1:
# event rates are incompatible
probG_ON = 0
probO_ON = 0
else:
probG_ON = 1
probO_ON = 1
if er != None and on != None:
probG_OM = math.sqrt(probG_ER * probG_ON)
probO_OM = math.sqrt(probO_ER * probO_ON)
elif er != None:
probG_OM = probG_ER
probO_OM = probO_ER
else:
# on != None
probG_OM = probG_ON
probO_OM = probO_ON
prob = round(probG_OM * probO_OM * probMultiplier)
else:
prob = 0
probMatrix[omIdx].append(prob)
# if sentence.abstract.id == '21600592':
# for omIdx in range(maxSize):
# for goIdx in range(maxSize):
# if omIdx < nOutcomeMeasurements and goIdx < nGroupOutcomePairs:
# om = outcomeMeasurements[omIdx]
# (group, outcome) = goPairs[goIdx]
# print probMatrix[omIdx][goIdx], om.statisticString(), group.name, outcome.name
costMatrix = munkres.make_cost_matrix(probMatrix, lambda cost: probMultiplier - cost)
m = munkres.Munkres()
# print probMatrix
# print costMatrix
indices = m.compute(costMatrix)
# threshold is (1/2)^4
threshold = 0.0625 * probMultiplier
threshold = 0.0001 * probMultiplier
# threshold = 0.25 * probMultiplier
for omIdx, goIdx in indices:
if omIdx < nOutcomeMeasurements and goIdx < nGroupOutcomePairs:
prob = probMatrix[omIdx][goIdx]
if prob > threshold:
# this quantity and mention should be associated
prob = float(prob) / probMultiplier
om = outcomeMeasurements[omIdx]
(group, outcome) = goPairs[goIdx]
self.linkOutcomeMeasurementAssociations(om, group, outcome, prob)
# record those outcome measurements that were not succefully matched to G,O
if omIdx < nOutcomeMeasurements and (goIdx >= nGroupOutcomePairs or probMatrix[omIdx][goIdx] <= threshold):
om = outcomeMeasurements[omIdx]
prob = float(probMatrix[omIdx][goIdx])/probMultiplier
if goIdx < nGroupOutcomePairs:
(group, outcome) = goPairs[goIdx]
else:
group = None
outcome = None
abstract = sentence.abstract
if abstract not in self.incompleteMatches:
self.incompleteMatches[abstract] = []
self.incompleteMatches[abstract].append(baseassociator.OutcomeMeasurementAssociation(group, outcome, om, prob))
def optimizeMatrix():
cost_matrix = make_cost_matrix(matrix, lambda cost: sys.maxint - cost)
m = Munkres()
indexes = m.compute(cost_matrix)
return indexes
def costify(similarity_matrix):
"""Transform a similarity matrix into a cost matrix."""
return munkres.make_cost_matrix(similarity_matrix, lambda s: 1 - s)
t_idx = np.argmax(score_matrix[s_idx])
if targets[s_idx, t_idx] > 0:
correct.append(s_idx)
else:
errors.append(s_idx)
total = len(correct) + len(errors)
print "Right: %d/%d = %f%%, Wrong: %d/%d = %f%%" \
% (len(correct), total, 100. * len(correct) / total,
len(errors), total, 100. * len(errors) / total)
sys.exit()
print "Finding best matching"
m = munkres.Munkres()
correct, errors = [], []
cost_matrix = munkres.make_cost_matrix(score_matrix, lambda cost: 1 - cost)
indexes = m.compute(cost_matrix)
for row, column in indexes:
if targets[row, column] > 0:
correct.append((row, column))
else:
errors.append((row, column))
total = len(correct) + len(errors)
print "Right: %d/%d = %f%%, Wrong: %d/%d = %f%%" \
% (len(correct), total, 100. * len(correct) / total,
len(errors), total, 100. * len(errors) / total)
# scores = cross_validation.cross_val_score(
# clf, m, targets, cv=5, scoring=scoring)
# print sum(predicted), sum(predicted - targets)
# print metrics.classification_report(targets, predicted)
gmm_model = sklmix.GMM(n_components=3, covariance_type='full') gmm_model.fit(iris[['PW', 'PL', 'SW']]) yhat = gmm_model.predict(iris[['PW', 'PL', 'SW']]) crosstab = pd.crosstab(iris['Type'], yhat, rownames=['true'], colnames=['predicted']) print crosstab # <headingcell level=4> # Align the confusion matrix with a non-standard package # <codecell> import munkres import sys m = munkres.Munkres() cost = munkres.make_cost_matrix(crosstab.values.tolist(), lambda x : sys.maxint - x) align = m.compute(cost) print align, '\n' permute = [x[1] for x in align] new_label = np.argsort(permute) yhat_new = new_label[yhat] print pd.crosstab(iris['Type'], yhat_new, rownames=['true'], colnames=['predicted']) # <headingcell level=4> # Bridging the gap with Rpy2 # <codecell> from rpy2.robjects import r
from munkres import Munkres, print_matrix, make_cost_matrix import sys n, m = map(int, raw_input().split()) matrix = [] for i in xrange(n): matrix.append(map(int, raw_input().split())) cost_matrix = make_cost_matrix(matrix, lambda cost: sys.maxint - cost) m = Munkres() indexes = m.compute(cost_matrix) for i, j in indexes: value = matrix[i][j] print '%d,%d' % (i, j)
def cost_matrix(contingency_table):
"""Hungarian method assumes the goal of minimum cost, while our goal with a
contigency table is maximum cost. To deal with this, the table is inverted in
an additive sense.
"""
return make_cost_matrix(contingency_table, lambda cost: sys.maxsize - cost)
def linkTemplatesHungarian(self, sentence):
""" link group size and group templates using Hungarian matching algorithm """
# print 'linking all templates'
templates = sentence.templates
costValueList = templates.getList('cost_value')
outcomeList = templates.getList('outcome')
groupList = templates.getList('group')
nCostValues = len(costValueList)
nGroupOutcomePairs = len(outcomeList)*len(groupList)
if nGroupOutcomePairs == 0 or nCostValues == 0:
return
maxSize = max(nCostValues, nGroupOutcomePairs)
# build list of group-outcome pairs
goPairs = []
for group in groupList:
for outcome in outcomeList:
goPairs.append((group, outcome))
# initialize cost matrix for matching cost values with group,outcome pairs
probMatrix = []
probMultiplier = 100000
for cvIdx in range(maxSize):
probMatrix.append([])
for goIdx in range(maxSize):
if cvIdx < nCostValues and goIdx < nGroupOutcomePairs:
cv = costValueList[cvIdx]
(group, outcome) = goPairs[goIdx]
outcomeFV = cv.getMatchFeatures(outcome)
groupFV = cv.getMatchFeatures(group)
groupProb = groupFV.prob
outcomeProb = outcomeFV.prob
prob = round(groupProb * outcomeProb * probMultiplier)
else:
# no association can be made
# this possible match involves either a dummy cost value or a dummy (group,outcome) pair
prob = 0
probMatrix[cvIdx].append(prob)
costMatrix = munkres.make_cost_matrix(probMatrix, lambda cost: probMultiplier - cost)
m = munkres.Munkres()
# print probMatrix
# print costMatrix
indices = m.compute(costMatrix)
# threshold is (1/2)^4
# threshold = 0.0625 * probMultiplier
threshold = 0.0001 * probMultiplier
# threshold = 0.25 * probMultiplier
for cvIdx, goIdx in indices:
if cvIdx < nCostValues and goIdx < nGroupOutcomePairs:
prob = probMatrix[cvIdx][goIdx]
if prob > threshold:
# this quantity and mention should be associated
prob = float(prob) / probMultiplier
om = costValueList[cvIdx]
(group, outcome) = goPairs[goIdx]
self.linkOutcomeMeasurementAssociations(om, group, outcome, prob)
