def run_real_ast_experiment(flag_savedata=True,genMethod='synthetic', nEst=20,nCand=80):
#parameters required
np.random.seed(1000)
price_range = 1000 #denotes highest possible price of a product
eps = 1 #tolerance
N = 1 # #number of times Monte Carlo simulation will run
if genMethod =='synthetic':
real_data_list = [
{'fname':'freq_itemset_data/retail0p0001_240852_txns88162.csv','isCSV':True,'min_ast_length':3},
{'fname':'freq_itemset_data/foodmartFIM0p0001_233231_txns4141.csv','isCSV':True,'min_ast_length':4},
{'fname':'freq_itemset_data/chains0p00001_txns1112949.txt','isCSV':False,'min_ast_length':5},
{'fname':'freq_itemset_data/OnlineRetail0p000001_txns540455.txt','isCSV':False,'min_ast_length':3}]
if genMethod == 'tafeng':
real_data_list = [{'fname':'freq_itemset_data/tafeng_final_0p00001_txns119390.txt','isCSV':False,'min_ast_length':8}]
algos = collections.OrderedDict({'Linear-Search':genAst_oracle,'Assort-LSH-G':genAst_AssortLSH,'Assort-Exact-G':genAst_AssortExact, 'Assort-BZ-G':genAst_AssortBZ})
benchmark = 'Linear-Search'
loggs = get_log_dict(real_data_list,N,algos,price_range,eps) #hack
loggs['additional']['real_data_list'] = real_data_list
badError = 0
t1= time.time()
for i,real_data in enumerate(real_data_list):
t0 = time.time()
t = 0
while(t<N):
print 'Iteration number is ', str(t+1),' of ',N,', for real ast data ',real_data['fname']
#generating the price
meta = {'eps':eps}
if genMethod =='synthetic':
p,v,feasibles,C,prod = generate_instance_general(price_range,None,'synthetic',t,lenFeas=None,real_data=real_data)
else:
p,v,feasibles,C,prod = generate_instance_general(price_range,None,'tafeng',t,lenFeas=None,real_data=real_data)
#
loggs['additional']['C'][i,t] = C
meta['feasibles'] = feasibles
#preprocessing for proposed algos
if 'Assort-Exact-G' in algos:
meta['db_exact'],_,meta['normConst'] = preprocess(prod, C, p, 'general_case_exact',feasibles=feasibles)
if 'Assort-LSH-G' in algos:
meta['db_LSH'],_,meta['normConst'] = preprocess(prod, C, p, 'general_case_LSH', nEst=nEst,nCand=nCand,feasibles=feasibles)#Hardcoded values
if 'Assort-BZ-G' in algos:
meta['db_BZ'],_,meta['normConst'] = preprocess(prod, C, p, 'general_case_BZ', nEst=nEst,nCand=nCand,feasibles=feasibles)#Hardcoded values
#run algos
maxSetBenchmark = None
for algoname in algos:
print '\tExecuting ',algoname
loggs[algoname]['rev'][i,t],loggs[algoname]['maxSet'][(i,t)],loggs[algoname]['time'][i,t] = algos[algoname](prod,C,p,v,meta)
print '\t\tTime taken is ',loggs[algoname]['time'][i,t],'sec.'
if algoname==benchmark:
maxSetBenchmark = copy.deepcopy(loggs[algoname]['maxSet'][(i,t)])
loggs,badError = compute_overlap_stats(benchmark,algos,loggs,i,t,badError,maxSetBenchmark,eps)
t = t+1
print 'Experiments (',N,' sims) for real ast data ',real_data['fname'], ' is done.'
print 'Cumulative time taken is', time.time() - t0,'\n'
loggs = compute_summary_stats(algos,loggs,benchmark,i)
#dump it incrementally for each product size
if flag_savedata == True:
pickle.dump(loggs,open('./output/gen_loggs_real_ast_upto'+str(i)+'_nCand_'+str(nCand)+'_nEst_'+str(nEst)+'_'+datetime.datetime.now().strftime("%Y%m%d_%I%M%p")+'.pkl','wb'))
print '\nAll experiments done. Total time taken is', time.time() - t1,'\n\n'
print "Summary:"
for algoname in algos:
print '\t',algoname,'time_mean',loggs[algoname]['time_mean']
print '\t',algoname,'revPctErr_mean',loggs[algoname]['revPctErr_mean']
return loggs
def run_prod_experiment_static_mnl(flag_capacitated=True,flag_savedata=True,genMethod='synthetic'):
#parameters required
random.seed(10)
np.random.seed(1000)
price_range = 1000 #denotes highest possible price of a product
eps = 0.1 #tolerance
N = 50 # #number of times Monte Carlo simulation will run
if flag_capacitated == True:
C = 100 #capacity of assortment
if genMethod=='synthetic':
prodList = [100, 250, 500, 1000] #[100,200,300] #
else:
prodList = [100, 250, 500, 1000]
algos = collections.OrderedDict({'Static-MNL':capAst_static_mnl, 'LP':capAst_LP}) #'LP':capAst_LP
benchmark = 'LP'
loggs = get_log_dict(prodList,N,algos,price_range,eps,C)
else:
prodList = [100,200,400,800,1600]
algos = collections.OrderedDict({'Linear-Search':genAst_oracle,'Assort-Exact-G':genAst_AssortExact,'Assort-LSH-G':genAst_AssortLSH})
benchmark = 'Linear-Search'
loggs = get_log_dict(prodList,N,algos,price_range,eps)
loggs['additional']['lenFeasibles'] = np.zeros(len(prodList))
badError = 0
t1= time.time()
for i,prod in enumerate(prodList):
t0 = time.time()
t = 0
while(t<N):
print 'Iteration number is ', str(t+1),' of ',N,', for prod size ',prod
#generating the price
meta = {'eps':eps}
if flag_capacitated == True:
if genMethod == 'tafeng':
file_1 = open("products_final.pkl",'rb')
product_choices = pickle.load(file_1)
choices = random.sample(product_choices,prod)
p,v = get_real_prices_parameters_by_product(choices)
else:
p,v = generate_instance(price_range,prod,genMethod,t)
else:
p,v,feasibles,C,prod = generate_instance_general(price_range,prod,genMethod,t)
loggs['additional']['C'][i,t] = C
meta['feasibles'] = feasibles
#preprocessing for proposed algos
if 'Assort-Exact' in algos:
meta['db_exact'],_,meta['normConst'] = preprocess(prod, C, p, 'special_case_exact')
if 'Assort-LSH' in algos:
meta['db_LSH'],_,_ = preprocess(prod, C, p, 'special_case_LSH', nEst=20,nCand=80)#Hardcoded values
if 'Assort-Exact-G' in algos:
meta['db_exact'],_,meta['normConst'] = preprocess(prod, C, p, 'general_case_exact',feasibles=feasibles)
if 'Assort-LSH-G' in algos:
meta['db_LSH'],_,_ = preprocess(prod, C, p, 'general_case_LSH', nEst=20,nCand=80,feasibles=feasibles)#Hardcoded values
#run algos
maxSetBenchmark = None
for algoname in algos:
print '\tExecuting ',algoname
loggs[algoname]['rev'][i,t],loggs[algoname]['maxSet'][(i,t)],loggs[algoname]['time'][i,t] = algos[algoname](prod,C,p,v,meta)
print '\t\tTime taken is ',loggs[algoname]['time'][i,t],'sec.'
if algoname==benchmark:
maxSetBenchmark = copy.deepcopy(loggs[algoname]['maxSet'][(i,t)])
loggs,badError = compute_overlap_stats(benchmark,algos,loggs,i,t,badError,maxSetBenchmark,eps)
t = t+1
print 'Experiments (',N,' sims) for number of products ',prod, ' is done.'
print 'Cumulative time taken is', time.time() - t0,'\n'
loggs = compute_summary_stats(algos,loggs,benchmark,i)
if flag_capacitated != True:
loggs['additional']['lenFeasibles'][i] = len(feasibles)
#dump it incrementally for each product size
if flag_savedata == True:
if flag_capacitated == True:
pickle.dump(loggs,open('./output/cap_loggs_'+genMethod+'_prod_'+str(prod)+'_'+datetime.datetime.now().strftime("%Y%m%d_%I%M%p")+'.pkl','wb'))
else:
pickle.dump(loggs,open('./output/gen_loggs_'+genMethod+'_prod_'+str(prod)+'_'+datetime.datetime.now().strftime("%Y%m%d_%I%M%p")+'.pkl','wb'))
print '\nAll experiments done. Total time taken is', time.time() - t1,'\n\n'
print "Summary:"
for algoname in algos:
print '\t',algoname,'time_mean',loggs[algoname]['time_mean']
print '\t',algoname,'revPctErr_mean',loggs[algoname]['revPctErr_mean']
return loggs
def run_lenFeas_experiment(flag_savedata=True,genMethod='synthetic',nEst=20,nCand=80):
#parameters required
random.seed(10)
np.random.seed(1000)
price_range = 1000 #denotes highest possible price of a product
eps = 1 #tolerance
N = 50 # #number of times Monte Carlo simulation will run
prod = 1000
lenFeasibles= [100,200,400,800,1600,3200,6400,12800,25600,51200]
#lenFeasibles= [51200]
algos = collections.OrderedDict({'Linear-Search':genAst_oracle, 'Assort-LSH-G':genAst_AssortLSH, 'Assort-Exact-G':genAst_AssortExact, 'Assort-BZ-G' : genAst_AssortBZ})
#algos = collections.OrderedDict({'Linear-Search':genAst_oracle,'Assort-LSH-G':genAst_AssortLSH,'Assort-Exact-G':genAst_AssortExact, 'Assort-BZ-G' : genAst_AssortBZ})
benchmark = 'Linear-Search'
loggs = get_log_dict(lenFeasibles,N,algos,price_range,eps) #hack
loggs['additional']['lenFeasibles'] = lenFeasibles
loggs['additional']['nEst'] = nEst
loggs['additional']['nCand'] = nCand
badError = 0
t1= time.time()
for i,lenFeas in enumerate(lenFeasibles):
t0 = time.time()
t = 0
while(t<N):
print 'Iteration number is ', str(t+1),' of ',N,', for no. of assortments ',lenFeas
#generating the price
meta = {'eps':eps}
p,v,feasibles,C,prod = generate_instance_general(price_range,prod,genMethod,t,lenFeas=lenFeas)
loggs['additional']['C'][i,t] = C
meta['feasibles'] = feasibles
#preprocessing for proposed algos
if 'Assort-Exact-G' in algos:
meta['db_exact'],_,meta['normConst'] = preprocess(prod, C, p, 'general_case_exact',feasibles=feasibles)
if 'Assort-LSH-G' in algos:
meta['db_LSH'],_,meta['normConst'] = preprocess(prod, C, p, 'general_case_LSH', nEst=nEst,nCand=nCand,feasibles=feasibles)#Hardcoded values
if 'Assort-BZ-G' in algos:
meta['db_BZ'],_,meta['normConst'] = preprocess(prod, C, p, 'general_case_BZ', nEst=nEst,nCand=nCand,feasibles=feasibles)#Hardcoded values
#run algos
maxSetBenchmark = None
for algoname in algos:
print '\tExecuting ',algoname
loggs[algoname]['rev'][i,t],loggs[algoname]['maxSet'][(i,t)],loggs[algoname]['time'][i,t] = algos[algoname](prod,C,p,v,meta)
print '\t\tTime taken is ',loggs[algoname]['time'][i,t],'sec.'
if algoname==benchmark:
maxSetBenchmark = copy.deepcopy(loggs[algoname]['maxSet'][(i,t)])
loggs,badError = compute_overlap_stats(benchmark,algos,loggs,i,t,badError,maxSetBenchmark,eps)
t = t+1
print 'Experiments (',N,' sims) for number of feasibles ',lenFeas, ' is done.'
print 'Cumulative time taken is', time.time() - t0,'\n'
loggs = compute_summary_stats(algos,loggs,benchmark,i)
#dump it incrementally for each product size
if flag_savedata == True:
pickle.dump(loggs,open('./output/gen_loggs_'+genMethod+'_lenF_'+str(lenFeas)+'_nCand_'+str(nCand)+'_nEst_'+str(nEst)+'_'+datetime.datetime.now().strftime("%Y%m%d_%I%M%p")+'.pkl','wb'))
print '\nAll experiments done. Total time taken is', time.time() - t1,'\n\n'
print "Summary:"
for algoname in algos:
print '\t',algoname,'time_mean',loggs[algoname]['time_mean']
print '\t',algoname,'revPctErr_mean',loggs[algoname]['revPctErr_mean']
return loggs