def method_1(self):
#creates a distance matrix that essentially is a list containing lists with the property where matrix[i][j] is the distance between point i and point j
t0 = dt.time()#starts to time the method until its completion
matrix = []
meth1sol = []
for i in li:
r = []
for j in li:
r.append(i.distance(j))
matrix.append(r)
n = len(matrix)
V = range(n)
E = [(i,j) for i in V for j in V if i!=j]
#the algorithm that eliminates invalid subtours
pm.begin('subtour elimination')
x = pm.var('x', E, bool)
#minimizes the sum of the distances found in the matrix
pm.minimize(sum(matrix[i][j]*x[i,j] for i,j in E), 'dist')
for k in V:
sum( x[k,j] for j in V if j!=k ) == 1
sum( x[i,k] for i in V if i!=k ) == 1
#calls the solver method and deactivates the result message
pm.solver(float, msg_lev = pm.glpk.GLP_MSG_OFF)
pm.solver(int, msg_lev= pm.glpk.GLP_MSG_OFF)
pm.solve()
global subtourg
#the function that creates subtours
def subtourl(x):
succ = 0
subt = [succ] #start from node 0
while True:
succ=sum(x[succ,j].primal*j for j in V if j!=succ)
if succ == 0: break #tour found
subt.append(int(succ+0.5))
return subt
subtourg = subtourl
while True:
#a loop that creates subtours and keeps them if they are valid, terminating the programme in the process, or discards them if they are not
subt = subtourg(x)
if len(subt) == n:
#print("Optimal tour length: %g"%pm.vobj())
#print("Optimal tour:"); print(subt)
break
print("New subtour: %r"% subt)
if len(subt) == 1: break #something wrong
#now add a subtour elimination constraint:
nots = [j for j in V if j not in subt]
sum(x[i,j] for i in subt for j in nots) >= 1
pm.solve() #solve the IP problem again
pm.end()
#print(subt)
#now the solution is added to a list that can be interpreted by the connect method
for i in subt:
meth1sol.append(li[i])
print(len(meth1sol))
self.connect(meth1sol, method1_colour, 1)
t1 = dt.time()
t = t1 - t0
t = round(t, 2)#the required time is calculated and rounded for conviniency
self.t1.set("time:\n{}s".format(t))
def predict(self, c, unknown):
if c.size < 2:
return []
addedElems = 0
if c.shape[1]-c.shape[0]>1:
added = numpy.zeros((c.shape[1]-c.shape[0], c.shape[1]))
addedElems = added.shape[0] * added.shape[1]
c = numpy.vstack((c, added))
detected = c.shape[0]
recognizers = c.shape[1]
size = c.size
mp.beginModel('basic')
mp.verbose(False)
x = mp.var(range(size), 'X', kind=bool)
# One label per detected object
for i in range(0, detected):
tmp = numpy.zeros((detected, recognizers))
tmp[i, :] = 1
tmp = tmp.reshape((1, size))[0]
mp.st(sum(x[j]*int(tmp[j]) for j in range(size))==1)
if self.constraints == True:
# Use recognizer up to once
if unknown==True:
lim = recognizers-1
else:
lim = recognizers
for i in range(0, recognizers):
tmp = numpy.zeros((detected, recognizers))
tmp[:, i] = 1
tmp = tmp.reshape((1, size))[0]
mp.st(sum(x[j]*int(tmp[j]) for j in range(size))<=1)
c = c.reshape((1, size)).tolist()[0]
mp.minimize(sum(c[i]*x[i] for i in range(size)), 'myobj')
mp.solve(int)
X = numpy.zeros((1, size))
for i in range(size):
X[0, i] = x[i].primal
X = X[0,0:size-addedElems]
labels = X.reshape(((size-addedElems)/(recognizers), recognizers))
predicted = []
for i in range(0, labels.shape[0]):
l = numpy.argmax(labels[i,:])
if l==recognizers-1 and unknown==True:
l = -1
predicted.append(l)
return predicted
vectorizer = TfidfVectorizer()
X = vectorizer.fit_transform(corpus)
print(vectorizer.get_feature_names())
print(X.shape)
print(X)
import pymprog as mp
mp.begin('bike production')
x, y = mp.var('x, y') # variablesas mp
mp.maximize(15 * x + 10 * y, 'profit')
x <= 3 # mountain bike limit
y <= 4 # racer production limit
x + y <= 5 # metal finishing limit
mp.solve()
print("#####################")
import spacy
import spacy_kenlm
nlp = spacy.load('en_core_web_sm')
kenlm_model = spacy_kenlm.spaCyKenLM(
'coca_fulltext.clean.lm.arpa') # default model from test.arpa
nlp.add_pipe(kenlm_model)
str = "I am a boy ."
doc = nlp(str)
#minOfferProductConstraint:
#for k in rNumberProducts:
# sum(x[i,j,k] for i in rNumberClients for j in rNumberChannels)\
# >=minOfferProduct[k]
#budgetConstraint:
pp.st(sum(x[i,j,k]*cost[j] for i in rNumberClients for j in rNumberChannels\
for k in rNumberProducts)<=budget, "Budget Constr")
#clientLimitConstraint:
for i in rNumberClients:
pp.st( sum(x[i,j,k] for j in rNumberChannels for k in rNumberProducts)<=1,
"client "+str(i)+" Limit")
#rurdleRateConstraint:
pp.st(sum(x[i,j,k]*expectedReturn[i][j][k] for i in rNumberClients for j in\
rNumberChannels for k in rNumberProducts)>= (1+rurdleRate)*sum(x[i,j,k]\
*cost[j] for i in rNumberClients for j in rNumberChannels\
for k in rNumberProducts),"rurdleRate Cons")
pp.solve() # solve the model
pp.sensitivity() # sensitivity report
print(pp.KKT())
print("Objetivo encontrado: ", round(pp.vobj(),2))
#pp.end() #Good habit: do away with the model
def ppSolver(expectedReturn, numberClients, numberChannels, numberProducts,
cost, budget, channelCap, minOfferProduct, maxOfferProduct,
rurdleRate):
startTime = timeit.default_timer()
rNumberClients = range(numberClients)
rNumberChannels = range(numberChannels)
rNumberProducts = range(numberProducts)
t = pp.iprod(rNumberClients, rNumberChannels, rNumberProducts)
pp.begin('basic') # begin modelling
pp.verbose(False) # be verbose
x = pp.var('choice', t, bool)
pp.maximize(sum(x[i,j,k]*expectedReturn[i][j][k] for i in rNumberClients\
for j in rNumberChannels for k in rNumberProducts))
#channelLimitConstraint:
for j in rNumberChannels:
sum(x[i,j,k] for i in rNumberClients for k in rNumberProducts)\
<=channelCap[j]
#maxOfferProductConstraint:
for k in rNumberProducts:
sum(x[i,j,k] for i in rNumberClients for j in rNumberChannels)\
<=maxOfferProduct[k]
#minOfferProductConstraint:
# for k in rNumberProducts:
# sum(x[i,j,k] for i in rNumberClients for j in rNumberChannels)\
# >=minOfferProduct[k]
#budgetConstraint:
pp.st(sum(x[i,j,k]*cost[j] for i in rNumberClients for j in\
rNumberChannels for k in rNumberProducts)<=budget,"Budget Constr.")
#clientLimitConstraint:
for i in rNumberClients:
pp.st(sum(x[i,j,k] for j in rNumberChannels for k in rNumberProducts)\
<=1,"Client "+str(i)+" limit")
#rurdleRateConstraint:
pp.st(sum(x[i,j,k]*expectedReturn[i][j][k] for i in rNumberClients for j \
in rNumberChannels for k in rNumberProducts)>= (1+rurdleRate)\
*sum(x[i,j,k]*cost[j] for i in rNumberClients for j in\
rNumberChannels for k in rNumberProducts),"Rurdle Rate Constr")
pp.solve() # solve the model
# pp.sensitivity() # sensitivity report
endTime = timeit.default_timer() - startTime
print("Objetivo encontrado: ", round(pp.vobj(), 2), " em ",
round(endTime, 3), " segundos")
print("\n\n\n")
appendCsv(numberClients, "Solver method", endTime, True,
round(pp.vobj(), 2))
pp.end() #Good habit: do away with the model
def merge ( Y, W ):
import operator
import sys
N = 4
M = 4
# index and data
patternid, yid, wid, rid = range(N*M), range(N*M), range(N*M), range(2)
#problem definition
pymprog.beginModel('basic')
pymprog.verbose(True)
# Stiamo facendo uno split, il pattern precedente e' P, e noi vogliamo creare due pattern figli Y e W
# P, Y, e W sono delle matrici NxM, rappresentate come vettori, di bit.
# N sono i layer, mentre M sono le hit in quel layer
P = pymprog.var(patternid, 'P', bool) # Parent Pattern
Y = pymprog.var(range(3), 'Y', bool) # First sub-pattern
W = pymprog.var(range(3), 'W', bool) # Second sub-pattenr
# Queste tre relazioni logiche impongono che per ogni traccia k, questa venga riconosciuta da almeno un sub-pattern
# e traducono la relazione
# covered[k] = coveredY[k] V coveredW[k] | k = {1,2,3}
# con
# covered[k] = 1 | k = {1,2,3}
covered = pymprog.var(range(3), 'covSubTot', bool)
r = pymprog.st( covered[k] >= Y[k] for k in range(3) )
r += pymprog.st( covered[k] >= W[k] for k in range(3) )
r += pymprog.st( covered[k] <= Y[k] + W[k] for k in range(3) )
r += pymprog.st( covered[k] == 1 for k in range(3) )
#
# # # # # # #
totY = pymprog.var( range(1), 'totY', bounds=(1, 3) )
totW = pymprog.var( range(1), 'totW', bounds=(1, 3) )
r += pymprog.st( sum ( Y[k] for k in range(3) ) == totY[i] for i in range(1) )
r += pymprog.st( sum ( W[k] for k in range(3) ) == totW[i] for i in range(1) )
AY = pymprog.var(range(N*M), 'AY', int)
AW = pymprog.var(range(N*M), 'AW', int)
## Ora voglio fare che *per ogni punto* nella griglia 2D,
r += pymprog.st ( AY[ i ] >= Y[k] * myTracks[k][ i ] for i in range(N*M) for k in range(3) )
r += pymprog.st ( AY[ i ] <= sum( Y[k] * myTracks[k][ i ] for k in range(3) ) for i in range(N*M) )
r += pymprog.st ( AW[ i ] >= W[k] * myTracks[k][ i ] for i in range(N*M) for k in range(3) )
r += pymprog.st ( AW[ i ] <= sum( W[k] * myTracks[k][ i ] for k in range(3) ) for i in range(N*M) )
# Le ampiezze nei vari layer ... poi si potra' semplificare
AmpY = pymprog.var(range(N), 'AmpY', int)
AmpW = pymprog.var(range(N), 'AmpW', int)
r += pymprog.st ( sum( AY[i*N+j] for j in mRange ) == AmpY[i] for i in nRange )
r += pymprog.st ( sum( AW[i*N+j] for j in mRange ) == AmpW[i] for i in nRange )
VolTotY = pymprog.var(range(1), 'VolTotY', int) # Second sub-pattenr
VolTotW = pymprog.var(range(1), 'VolTotW', int) # Second sub-pattenr
#r += st ( sum( AY[i*N+j] for j in mRange ) == AmpY[i] for i in nRange )
r += pymprog.st ( sum( AmpY[j] for j in nRange ) == VolTotY[i] for i in range(1) )
r += pymprog.st ( sum( AmpW[j] for j in nRange ) == VolTotW[i] for i in range(1) )
pymprog.minimize( VolTotY[0] + VolTotW[0] , 'Total Volume')
sys.stdout.write("\nSolving ...")
pymprog.solve()
sys.stdout.write(" done.\n\n")
print("Total Volume = %g"% pymprog.vobj())
print 'Y'
print Y
print 'W'
print W
print AmpY
print AmpW
print 'AY'
print_variables_matrix_primal(AY)
print 'AW'
print_variables_matrix_primal(AW)
print 'Y'
print_variables_matrix_cross(AY)
print 'W'
print_variables_matrix_cross(AW)
print 'Volume Y : '
print VolTotY[0].primal
print 'Volume W : '
print VolTotW[0].primal
print 'Volume Tot : '
print sum( tracks[i] for i in range(N*M) )
def summerize(tweets_df):
print(len(tweets_df))
#print(tweets_df['tweet_texts'][1])
tf_idf.compute_tf_idf(tweets_df)
term_matrix = np.load('term_matrix.npy')
vocab_to_idx = np.load('vocab_to_idx.npy', allow_pickle=True).item()
content_vocab = list(np.load('content_vocab.npy'))
# tfidf_dict = np.load('tfidf_dict.npy', allow_pickle=True).item()
print("1 ##################")
spacy_tweets = []
for doc in nlp.pipe(tweets_df['tweet_texts'].astype('unicode'),
n_threads=-1):
spacy_tweets.append(doc)
spacy_tweets = [tweet for tweet in spacy_tweets if len(tweet) > 1]
# spacy_tweets = np.random.choice(spacy_tweets, 10, replace=False)
# spacy_tweets = spacy_tweets[:20]
print(len(spacy_tweets))
print(spacy_tweets[0])
print("2 ##################")
all_bigrams = [
list(bigrams([token.lemma_ for token in tweets]))
for tweets in spacy_tweets
]
starting_nodes = [single_bigram[0] for single_bigram in all_bigrams]
end_nodes = [single_bigram[-1] for single_bigram in all_bigrams]
all_bigrams = [
node for single_bigram in all_bigrams for node in single_bigram
]
all_bigrams = list(set(all_bigrams))
print("all_bigrams len=", len(all_bigrams))
print(all_bigrams[0])
print("3 ##################")
# bigram_graph = make_bigram_graph(all_bigrams, starting_nodes[1])
# print(len(bigram_graph))
# print(bigram_graph)
# path = breadth_first_search(bigram_graph, starting_nodes[1], end_nodes[2])
# print(path)
bigram_paths = []
for single_start_node in tqdm(starting_nodes):
bigram_graph = make_bigram_graph(all_bigrams, single_start_node)
for single_end_node in end_nodes:
possible_paths = breadth_first_search(bigram_graph,
single_start_node,
single_end_node)
for path in possible_paths:
bigram_paths.append(path)
print("bigram_paths len=", len(bigram_paths))
# print(bigram_paths[10])
# for tweet in spacy_tweets:
# bigram_paths.append(list(bigrams([token.lemma_ for token in tweets])))
word_paths = []
for path in tqdm(bigram_paths):
word_paths.append(make_list(path))
print(word_paths[0])
print("4 ##################")
mp.begin('COWABS')
# Defining my first variable, x
# This defines whether or not a word path is selected
x = mp.var(str('x'), len(word_paths), bool)
# Also defining the second variable, which defines
# whether or not a content word is chosen
y = mp.var(str('y'), len(content_vocab), bool)
mp.maximize(
sum([
linguistic_quality(word_paths[i]) *
informativeness(word_paths[i], term_matrix, vocab_to_idx) * x[i]
for i in range(len(x))
]) + sum(y))
# hiding the output of this line since its a very long sum
# sum([x[i] * len(word_paths[i]) for i in range(len(x))]) <= 150
for j in range(len(y)):
sum([
x[i]
for i in paths_with_content_words(j, word_paths, content_vocab)
]) >= y[j]
for i in range(len(x)):
sum(y[j] for j in content_words(i, word_paths, content_vocab)) >= len(
content_words(i, word_paths, content_vocab)) * x[i]
mp.solve()
result_x = [value.primal for value in x]
result_y = [value.primal for value in y]
mp.end()
chosen_paths = np.nonzero(result_x)
chosen_words = np.nonzero(result_y)
print("*** Total = ", len(chosen_paths[0]))
min_cosine_sim = 999
final_sentence = None
for i in chosen_paths[0]:
print('--------------')
print(str(" ").join([token for token in word_paths[i]]))
cosine_sim = informativeness(word_paths[i], term_matrix, vocab_to_idx)
print(cosine_sim)
if min_cosine_sim > cosine_sim:
min_cosine_sim = cosine_sim
final_sentence = str(" ").join([token for token in word_paths[i]])
# print("####### Summary ###########")
# print(final_sentence)
return final_sentence
def solve_offline(self, w):
''' pre-compute offline solution '''
T = range(self.T)
self.w = dict.fromkeys(T, 0) # vector of charges and discharges
# making some shortcuts so the LP is more readable and similar to
# the model in our paper
b0 = self.level
# we use a positive Bm in this problem (like d)
Bp = Bm = self.max_rate
B = self.capacity
CC = self.street.C
alpha = self.efficiency
p = self.exp_prices
ka = self.kwh_adj
# magnitudes without me involved
fp0 = [self.street.f(t, 0, p[t])[0] for t in T]
fm0 = [self.street.f(t, 0, p[t])[1] for t in T]
# maximal limit for magnitude we expect
maxx = 100
# maximal number of consecutive critical steps considered
K = range(self.max_k)
e = [math.pow(self.c_h, k + 1) for k in K]
TK = [(t,k) for t in T for k in K if k <= t]
pymprog.beginModel('offline')
#pymprog.verbose(True)
pymprog.solvopt(tm_lim=1000*60*self.max_runtime)
#print "Options:", pymprog.solvopt()
# create variables
c = pymprog.var(T, 'C') # charges
d = pymprog.var(T, 'D') # discharges
x = pymprog.var(T, 'X') # max mag on cable
# True if there have been at least k-1 consecutive
# critical steps before t
v = pymprog.var(T, 'V') # cost
co = pymprog.var(TK, 'CO', bool) # consecutive overload
# set objective
pymprog.minimize(
w * sum(v[t] for t in T) - sum(p[t] * (d[t] - c[t]) for t in T),
'wC-R'
# we use a positive d in this problem
)
# set constraints
# ... battery contraints
pymprog.st(0 <= c[t] <= Bp for t in T)
pymprog.st(0 <= d[t] <= Bm for t in T)
# we use a positive d in this problem
for t in T:
pymprog.st(b0 + sum(ka * (alpha * c[j]) - ka * d[j] for j in range(t + 1)) <= B)
pymprog.st(b0 + sum(ka * (alpha * c[j]) - ka * d[j] for j in range(t + 1)) >= 0)
# ... cable constraints
pymprog.st(x[t] >= 0 for t in T)
pymprog.st(x[t] >= fp0[t] + c[t] - d[t] for t in T)
pymprog.st(x[t] >= -fm0[t] - c[t] + d[t] for t in T)
pymprog.st(x[t] >= -fp0[t] - c[t] + d[t] for t in T)
pymprog.st(x[t] >= fm0[t] + c[t] - d[t] for t in T)
# ... cons. overloading constraints
pymprog.st(v[t] >= e[k] * co[t, k] for t, k in TK)
pymprog.st(co[t, 0] >= (x[t]-CC)/(maxx - CC) for t in T)
pymprog.st(co[t, k] >= co[t-1, k-1] + co[t, k-1] - 1.5 for t, k in TK if ((t, k-1) in TK and (t-1, k-1) in TK))
pymprog.solve()
# status can be in (opt, feas, undef)
assert(pymprog.status() in ('opt', 'feas'))
# value of objective (this should meet the actual outcome!)
#print 'Objective = {}'.format(pymprog.vobj())
# use sthg like this to inspect variables
#print ';\n'.join('%s = %g {dual: %g}' % (
# x[t].name, x[t].primal, x[t].dual)
# for t in T)
for t in T:
self.w[t] = c[t].primal - d[t].primal