def run_ours(Xtr,
Ytr,
Xt,
Yt,
lb,
nsample,
lambda_mode,
q,
sample_mode,
k=None,
rerun=True,
eps=0.01,
min_recall_per_class=0.8,
log=None):
#name = 'ours' if k is None else 'oursk'
name = 'ours{}'.format(int(rerun))
k = k if k is not None else 100
dec = DecisionSet(eps)
dec.train(Xtr,
Ytr,
max_k=k,
nsamp=nsample,
lamb=lambda_mode,
q=q,
mode=sample_mode,
rerun=rerun,
min_recall_per_class=min_recall_per_class)
print('default:', dec.default)
Xt_ = [Transaction(feat2item(t)) for t in Xt.values]
Y_pred = dec.predict_all(Xt_)
if log is None:
from logger import log
log('{}-default'.format(name), dec.default)
log('{}-k'.format(name), len(dec.rules))
log('{}-maxk'.format(name), k)
[log('{}-nconds'.format(name), len(r), i) for i, r in enumerate(dec.rules)]
log('{}-q'.format(name), q)
log('{}-nsample'.format(name), nsample)
log('{}-lamb'.format(name), lambda_mode)
log('{}-seq'.format(name), dec.seq)
log('{}-auc'.format(name),
roc_auc_score(lb.transform(Yt.values), lb.transform(Y_pred)))
log('{}-bacc'.format(name), balanced_accuracy_score(Yt, Y_pred))
log('{}-disp'.format(name), dispersion(dec.rules, average=True))
log('{}-overlap'.format(name), overlap(dec.rules))
log('{}-mode'.format(name), sample_mode)
[
log('{}-precisions-tr'.format(name), v, l)
for l, v in precision(dec).items()
]
[
log('{}-recall-tr'.format(name), v, l)
for l, v in recall(dec.rules).items()
]
print(confusion_matrix(Yt, Y_pred))
return Y_pred
def evaluate_mutator(mutator, threshold, min_percent, latex=True):
labels = (POS, NEU, NEG)
train_loc = root+'Data/twitterData/train_alternative.tsv'
dev_loc = root+'Data/twitterData/dev_alternative.tsv'
test_loc = root+'Data/twitterData/test_alternative.tsv'
train, dev, test = get_final_semeval_data(reduce(lambda x, y: x|y, labels), train_loc, dev_loc, test_loc)
dev_x, dev_y = dev
for label in labels:
pred_y = []
for tweet in dev_x:
pred_y.append(label if mutator.apply_filter(tweet, label) else -1)
if pred_y.count(label) < min_percent*len(pred_y):
yield str(label), ' (%.3f,0.0)' % threshold
else:
yield str(label), ' (%.3f,%.4f)' % (threshold, precision(dev_y, pred_y, label))
def calculateLSTMaccuracy(receipts, results):
total_price_total = 0
total_price_found = 0
total_price_correct = 0
currency_total = 0
currency_found = 0
currency_correct = 0
date_total = 0
date_found = 0
date_correct = 0
vendor_total = 0
vendor_found = 0
vendor_correct = 0
tax_rate_total = 0
tax_rate_found = 0
tax_rate_correct = 0
address_total = 0
address_found = 0
address_correct = 0
products_total = 0
products_found = 0
products_correct = 0
count = 0
for i, receipt in enumerate(receipts):
corr = True
## Check total price
if 'total_price' in results[i]:
price = results[i]['total_price'].replace(',','.')
to_remove = []
for p in price:
if util.isInt(p) or p == '.':
continue
to_remove.append(p)
for p in to_remove:
price = price.replace(p, '')
if price.count('.') == 2:
index = price.index('.')
price = price[0 : index : ] + price[index + 1 : :]
elif price.count('.') == 1 and len(price.split('.')[-1]) > 2:
price = price.replace('.', '')
else:
price = None
if price and price != '':
total_price_found+=1
if 'total_price' in receipt.groundTruth:
total_price_total+= 1
if compare.totalPrice(receipt.groundTruth['total_price'], price):
total_price_correct += 1
else:
corr = False
## Check currecy
if 'currency' in results[i]:
currency = results[i]['currency']
to_remove = []
for c in currency:
if c.isalpha():
continue
to_remove.append(c)
for c in to_remove:
currency = currency.replace(c, '')
else:
currency = None
if currency and currency != '':
currency_found+=1
if 'currency' in receipt.groundTruth:
currency_total+=1
if compare.currency(receipt.groundTruth['currency'], currency):
currency_correct += 1
else:
corr = False
## Check date
if 'date' in results[i]:
date = results[i]['date']
split = date.split(' ')
if len(split) == 2:
date = split[0]
if len(split[1]) > len(split[0]):
date = split[1]
else:
date = None
if date and date != '':
date_found+=1
if 'date' in receipt.groundTruth:
date_total+=1
if compare.date(receipt.groundTruth['date'],date):
date_correct += 1
else:
corr = False
## Check vendor
if 'vendor' in results[i]:
vendor = results[i]['vendor']
else:
vendor = None
if vendor and vendor != '':
vendor_found +=1
if 'vendor' in receipt.groundTruth:
vendor_total+=1
if compare.vendor(receipt.groundTruth['vendor'], vendor):
vendor_correct += 1
else:
corr = False
## Check tax rate
if 'tax_rate' in results[i]:
tax = results[i]['tax_rate']
split = tax.split(' ')
if len(split) == 2:
tax = split[0]
else:
tax = None
if tax and tax != '':
tax_rate_found+=1
if 'tax_rate' in receipt.groundTruth:
tax_rate_total+=1
if compare.taxRate(receipt.groundTruth['tax_rate'], tax):
tax_rate_correct += 1
else:
corr = False
## Check address
if 'address' in results[i]:
address = results[i]['address']
else:
address = None
if address and address != '':
address_found += 1
if 'address' in receipt.groundTruth:
address_total+=1
if compare.address(receipt.groundTruth['address'], address):
address_correct += 1
else:
corr = False
if 'products' in receipt.groundTruth:
products_total += len(receipt.groundTruth['products'])
if 'products' in results[i]:
products = results[i]['products']
found = []
for product in products:
product['amount'] = 1
products_found += 1
if not 'name' in product:
continue
if 'products' in receipt.groundTruth:
real_products = receipt.groundTruth['products']
for j,real_product in enumerate(real_products):
if j in found:
continue
if compare.products(product, real_product):
found.append(j)
products_correct += 1
break
totalDataPoints = vendor_total + date_total + address_total + tax_rate_total + total_price_total + currency_total + products_total
totalDataPointsFound = vendor_found + date_found + address_found + tax_rate_found + total_price_found + currency_found + products_found
totalCorrect = vendor_correct + date_correct + address_correct + tax_rate_correct + total_price_correct + currency_correct + products_correct
total_precision = 0
total_recall = 0
print('-----TOTAL CORRECT RECEIPTS-----')
print(count, 'of', len(receipts))
print('-----VENDORS-----')
print(vendor_total, vendor_found, vendor_correct)
precision = util.precision(vendor_correct, vendor_found)
recall = util.recall(vendor_total, vendor_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----DATES-----')
print(date_total, date_found, date_correct)
precision = util.precision(date_correct, date_found)
recall = util.recall(date_total, date_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----ADDRESSES-----')
print(address_total, address_found, address_correct)
precision = util.precision(address_correct, address_found)
recall = util.recall(address_total, address_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----TAX RATES-----')
print(tax_rate_total, tax_rate_found, tax_rate_correct)
precision = util.precision(tax_rate_correct, tax_rate_found)
recall = util.recall(tax_rate_total, tax_rate_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----PRICE-----')
print(total_price_total, total_price_found, total_price_correct)
precision = util.precision(total_price_correct, total_price_found)
recall = util.recall(total_price_total, total_price_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----CURRENCY-----')
print(currency_total, currency_found, currency_correct)
precision = util.precision(currency_correct, currency_found)
recall = util.recall(currency_total, currency_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----PRODUCTS-----')
print(products_total, products_found, products_correct)
precision = util.precision(products_correct, products_found)
recall = util.recall(products_total, products_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----MICRO AVG-----')
print(totalDataPoints, totalDataPointsFound, totalCorrect)
precision = util.precision(totalCorrect, totalDataPointsFound)
recall = util.recall(totalDataPoints, totalCorrect)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----MACRO AVG-----')
print(totalDataPoints, totalDataPointsFound, totalCorrect)
precision = total_precision / 7.0
recall = total_recall / 7.0
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
def calculateRuleBasedAccuracy(receipts):
total_price_total = 0
total_price_found = 0
total_price_correct = 0
currency_total = 0
currency_found = 0
currency_correct = 0
date_total = 0
date_found = 0
date_correct = 0
vendor_total = 0
vendor_found = 0
vendor_correct = 0
tax_rate_total = 0
tax_rate_found = 0
tax_rate_correct = 0
address_total = 0
address_found = 0
address_correct = 0
products_total = 0
products_found = 0
products_correct = 0
count = 0
for receipt in receipts:
corr = True
## Check total price
if 'total_price' in receipt.ruleBasedPrediction:
price = receipt.ruleBasedPrediction['total_price']
else:
price = None
if price:
total_price_found+=1
if 'total_price' in receipt.groundTruth:
total_price_total+= 1
if compare.totalPrice(receipt.groundTruth['total_price'], price):
total_price_correct += 1
else:
corr = False
## Check currecy
if 'currency' in receipt.ruleBasedPrediction:
currency = receipt.ruleBasedPrediction['currency']
else:
currency = None
if currency:
currency_found+=1
if 'currency' in receipt.groundTruth:
currency_total+=1
if compare.currency(receipt.groundTruth['currency'], currency):
currency_correct += 1
else:
corr = False
## Check date
if 'date' in receipt.ruleBasedPrediction:
date = receipt.ruleBasedPrediction['date']
else:
date = None
if date:
date_found+=1
if 'date' in receipt.groundTruth:
date_total+=1
if compare.date(receipt.groundTruth['date'],date):
date_correct += 1
else:
corr = False
## Check vendor
if 'vendor' in receipt.ruleBasedPrediction:
vendor = receipt.ruleBasedPrediction['vendor']
else:
vendor = None
if vendor:
vendor_found +=1
if 'vendor' in receipt.groundTruth:
vendor_total+=1
if compare.vendor(receipt.groundTruth['vendor'], vendor):
vendor_correct += 1
else:
corr = False
## Check tax rate
if 'tax_rate' in receipt.ruleBasedPrediction:
tax = receipt.ruleBasedPrediction['tax_rate']
else:
tax = None
if tax:
tax_rate_found+=1
if 'tax_rate' in receipt.groundTruth:
tax_rate_total+=1
if compare.taxRate(receipt.groundTruth['tax_rate'], tax):
tax_rate_correct += 1
else:
corr = False
## Check address
if 'address' in receipt.ruleBasedPrediction:
address = receipt.ruleBasedPrediction['address']
else:
address = None
if address:
address_found += 1
if 'address' in receipt.groundTruth:
address_total+=1
if compare.address(receipt.groundTruth['address'], address):
address_correct += 1
else:
corr = False
## Check products
if 'products' in receipt.ruleBasedPrediction:
products = receipt.ruleBasedPrediction['products']
else:
products = []
found = []
if 'products' in receipt.groundTruth:
products_total+= len(receipt.groundTruth['products'])
for product in products:
products_found += 1
if 'products' in receipt.groundTruth:
real_products = receipt.groundTruth['products']
for j,real_product in enumerate(real_products):
if j in found:
continue
if compare.products(product, real_product):
found.append(j)
products_correct += 1
break
if 'products' in receipt.groundTruth:
if len(found) < len(receipt.groundTruth['products']):
corr = False
if corr:
count+=1
totalDataPoints = vendor_total + date_total + address_total + tax_rate_total + total_price_total + currency_total + products_total
totalDataPointsFound = vendor_found + date_found + address_found + tax_rate_found + total_price_found + currency_found + products_found
totalCorrect = vendor_correct + date_correct + address_correct + tax_rate_correct + total_price_correct + currency_correct + products_correct
total_precision = 0
total_recall = 0
print('-----TOTAL CORRECT RECEIPTS-----')
print(count, 'of', len(receipts))
print('-----VENDORS-----')
print(vendor_total, vendor_found, vendor_correct)
precision = util.precision(vendor_correct, vendor_found)
recall = util.recall(vendor_total, vendor_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----DATES-----')
print(date_total, date_found, date_correct)
precision = util.precision(date_correct, date_found)
recall = util.recall(date_total, date_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----ADDRESSES-----')
print(address_total, address_found, address_correct)
precision = util.precision(address_correct, address_found)
recall = util.recall(address_total, address_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----TAX RATES-----')
print(tax_rate_total, tax_rate_found, tax_rate_correct)
precision = util.precision(tax_rate_correct, tax_rate_found)
recall = util.recall(tax_rate_total, tax_rate_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----PRICE-----')
print(total_price_total, total_price_found, total_price_correct)
precision = util.precision(total_price_correct, total_price_found)
recall = util.recall(total_price_total, total_price_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----CURRENCY-----')
print(currency_total, currency_found, currency_correct)
precision = util.precision(currency_correct, currency_found)
recall = util.recall(currency_total, currency_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----PRODUCTS-----')
print(products_total, products_found, products_correct)
precision = util.precision(products_correct, products_found)
recall = util.recall(products_total, products_correct)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----MICRO AVG-----')
print(totalDataPoints, totalDataPointsFound, totalCorrect)
precision = util.precision(totalCorrect, totalDataPointsFound)
recall = util.recall(totalDataPoints, totalCorrect)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----MACRO AVG-----')
print(totalDataPoints, totalDataPointsFound, totalCorrect)
precision = total_precision / 7.0
recall = total_recall / 7.0
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
def kmeans(eac, removeTerms, ngram):
terms = ['lazaro', 'lázaro', 'baez', 'báez', 'carlitos']
print('Filtering tweets')
tweets = util.read_from_file("dataset.csv")
if removeTerms:
tweets = filter.filter_tweets(tweets, terms_to_remove=terms)
else:
tweets = filter.filter_tweets(tweets)
# Reduce tweets list length
tweets = tweets[0:6300]
carlitos = 0
lazaro = 0
data = []
for tw in tweets:
if tw.tw_type == 'Carlitos':
carlitos += 1
else:
lazaro += 1
data.append(tw.text)
print(carlitos, lazaro)
print("Transform Data...")
# Transform data
if ngram:
hasher = HashingVectorizer(non_negative=True,
ngram_range=(1, 3),
analyzer='word',
norm='l2',
binary=False)
else:
hasher = HashingVectorizer(non_negative=True, norm='l2', binary=False)
vectorizer = make_pipeline(hasher)
X = vectorizer.fit_transform(data)
count = 0
precision_list = []
while count < 100:
# Start timer
t0 = time()
if eac:
clustering = EAC(30, min_k=2, max_k=10)
EAC_D = clustering.fit(X).distance_
# Kmedoids over EAC_D
kmed = KMedoids(2, init='random', distance_metric="precomputed")
labels = kmed.fit(EAC_D).labels_
else:
km = KMeans(n_clusters=2, init='k-means++', n_init=1, max_iter=100)
labels = km.fit(X).labels_
# Assign labels to tweets
for i in range(len(tweets)):
tweets[i].label = labels[i]
print("Precision: ")
# Print precision
precision = util.precision(tweets)
print("done in %0.3fs" % (time() - t0))
if isAdable(precision):
precision_list.append(precision)
count += 1
return precision_list
def minhash(eac, shingle, removeTerms):
terms = ['lazaro', 'lázaro', 'baez', 'báez', 'carlitos']
print('Filtering tweets')
tweets = util.read_from_file("dataset.csv")
if removeTerms:
tweets = filter.filter_tweets(tweets, terms_to_remove=terms)
else:
tweets = filter.filter_tweets(tweets)
# Reduce tweets list length
tweets = tweets[0:6300]
carlitos = 0
lazaro = 0
data = []
for tw in tweets:
if tw.tw_type == 'Carlitos':
carlitos += 1
else:
lazaro += 1
data.append(tw.text)
print(carlitos, lazaro)
# Extract text from tweets
X = [tw.text for tw in tweets]
# Start timer
t0 = time()
print("Calculating distance matrix...")
D = metrics.jaccard_minhash_distance_mp(X, shingle_length=shingle)
count = 0
precision_list = []
while count < 100:
if eac:
print("EAC clustering...")
# EAC clustering
kmedoid = KMedoids(init='random', distance_metric='precomputed')
clustering = EAC(30, min_k=2, max_k=10, clustering=kmedoid)
EAC_D = clustering.fit(D).distance_
# Kmedoids over EAC_D
kmed = KMedoids(2, init='random', distance_metric="precomputed")
labels = kmed.fit(EAC_D).labels_
else:
kmedoid = KMedoids(2, init='random', distance_metric='precomputed')
print("Kmedoids clustering...")
labels = kmedoid.fit(D).labels_
# Assign labels to tweets
for i in range(len(tweets)):
tweets[i].label = labels[i]
# Print precision
print("Precision: ")
precision = util.precision(tweets)
if isAdable(precision):
print(count)
precision_list.append(precision)
count += 1
print("done in %0.3fs" % (time() - t0))
return precision_list
def calculateMetrics(reciepts, result, writeToFile=False, path=None):
correctVendors = 0
vendorsFound = 0
vendors = 0
correctDates = 0
datesFound = 0
dates = 0
correctAddresses = 0
addressesFound = 0
addresses = 0
correctTaxes = 0
taxesFound = 0
taxes = 0
correctPrices = 0
pricesFound = 0
prices = 0
correctCurrencies = 0
currenciesFound = 0
currencies = 0
correctProducts = 0
productsFound = 0
products = 0
result_dict = {}
count = 0
for i, reciept in enumerate(reciepts):
corr = True
vendor = result[i]['vendor']
result_dict['vendor'] = vendor
if vendor:
vendorsFound += 1
vendor = vendor.lower()
if 'vendor' in reciept.groundTruth:
vendors += 1
if vendor and levenshtein_distance(
vendor, reciept.groundTruth['vendor'].lower()) <= 0:
correctVendors += 1
else:
corr = False
date = result[i]['date']
result_dict['date'] = date
if date:
datesFound += 1
date = date.lower()
if 'date' in reciept.groundTruth:
dates += 1
if date == reciept.groundTruth['date'].lower(
) or date == reciept.groundTruth['date'].lower().replace(' ', ''):
correctDates += 1
else:
corr = False
address = result[i]['address']
result_dict['address'] = address
if address:
addressesFound += 1
address = address.lower()
if 'address' in reciept.groundTruth:
addresses += 1
if address and levenshtein_distance(
address, reciept.groundTruth['address'].lower()) <= 0:
correctAddresses += 1
else:
corr = False
tax = result[i]['tax_rate']
result_dict['tax_rate'] = tax
if tax != None:
taxesFound += 1
if 'tax_rate' in reciept.groundTruth:
taxes += 1
real_tax = int(
float(reciept.groundTruth['tax_rate'].lower().replace('%',
'')))
if tax == real_tax:
correctTaxes += 1
else:
corr = False
price = result[i]['total_price']
result_dict['total_price'] = price
if price:
pricesFound += 1
if 'total_price' in reciept.groundTruth:
prices += 1
real_price = float(reciept.groundTruth['total_price'].lower())
if price == real_price:
correctPrices += 1
else:
corr = False
currency = result[i]['currency']
result_dict['currency'] = currency
if currency:
currenciesFound += 1
currency = currency.lower()
if 'currency' in reciept.groundTruth:
currencies += 1
if currency == reciept.groundTruth['currency'].lower():
correctCurrencies += 1
else:
corr = False
productsList = result[i]['products']
result_dict['products'] = productsList
if 'products' in reciept.groundTruth:
for product in reciept.groundTruth['products']:
products += 1
checkedIndexes = []
for product in productsList:
productsFound += 1
for i, real_product in enumerate(reciept.groundTruth['products']):
if i in checkedIndexes:
continue
price = None
if 'price' in product:
price = product['price'].replace(',', '.')
try:
price = float(price)
except:
price = None
real_price = real_product['price']
real_price = float(real_price)
if levenshtein_distance(product['name'].lower(),
real_product['name'].lower()) <= 0:
if util.floatCompare(price, real_price):
if product['amount'] == real_product['amount']:
correctProducts += 1
checkedIndexes.append(i)
break
if len(checkedIndexes) < len(reciept.groundTruth['products']):
corr |= False
if corr:
count += 1
if writeToFile:
with open(os.path.join(path, reciept.path), 'w') as fp:
json.dump(result_dict, fp, indent=1)
totalDataPoints = vendors + dates + addresses + taxes + prices + currencies + products
totalDataPointsFound = vendorsFound + datesFound + addressesFound + taxesFound + pricesFound + currenciesFound + productsFound
totalCorrect = correctVendors + correctDates + correctAddresses + correctTaxes + correctPrices + correctCurrencies + correctProducts
total_precision = 0
total_recall = 0
print('-----TOTAL CORRECT RECEIPTS-----')
print(count, 'of', len(reciepts))
print('-----VENDORS-----')
print(vendors, vendorsFound, correctVendors)
precision = util.precision(correctVendors, vendorsFound)
recall = util.recall(vendors, correctVendors)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----DATES-----')
print(dates, datesFound, correctDates)
precision = util.precision(correctDates, datesFound)
recall = util.recall(dates, correctDates)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----ADDRESSES-----')
print(addresses, addressesFound, correctAddresses)
precision = util.precision(correctAddresses, addressesFound)
recall = util.recall(addresses, correctAddresses)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----TAX RATES-----')
print(taxes, taxesFound, correctTaxes)
precision = util.precision(correctTaxes, taxesFound)
recall = util.recall(taxes, correctTaxes)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----PRICE-----')
print(prices, pricesFound, correctPrices)
precision = util.precision(correctPrices, pricesFound)
recall = util.recall(prices, correctPrices)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----CURRENCY-----')
print(currencies, currenciesFound, correctCurrencies)
precision = util.precision(correctCurrencies, currenciesFound)
recall = util.recall(currencies, correctCurrencies)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----PRODUCTS-----')
print(products, productsFound, correctProducts)
precision = util.precision(correctProducts, productsFound)
recall = util.recall(products, correctProducts)
total_precision += precision
total_recall += recall
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----MICRO AVG-----')
print(totalDataPoints, totalDataPointsFound, totalCorrect)
precision = util.precision(totalCorrect, totalDataPointsFound)
recall = util.recall(totalDataPoints, totalCorrect)
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
print('-----MACRO AVG-----')
print(totalDataPoints, totalDataPointsFound, totalCorrect)
precision = total_precision / 7.0
recall = total_recall / 7.0
print('Precision:', precision)
print('Recall:', recall)
print('F1:', util.fScore(precision, recall))
def train(self,
X,
Y,
max_k=100,
nsamp=100,
lamb=None,
q='kl',
mode=3,
rerun=True,
min_recall_per_class=0.5):
print('##### START #####')
Itemset.clear_db()
prep_db(X, Y)
# Allow specify lamb to a certain number by users
if type(lamb) == str or lamb is None:
samp = self.sample_from_each_label(set(Itemset.labels), 100, set(),
mode)
if lamb == 'max':
lamb = np.max([Rule.quality([r], metric=q) for r in samp])
elif lamb == 'mean':
lamb = np.mean([Rule.quality([r], metric=q) for r in samp])
else:
lamb = 0
print('lamb:', lamb)
greed = GreedyDiv([], lamb)
U_all = []
labels_samp = set(Itemset.labels)
while len(self) < max_k and len(labels_samp) > 0:
if mode == 0:
samps = []
for label in labels_samp:
_, samp = sample_rn(nsamp, label)
samp = [Rule(s, label)
for s in list(samp)] # Very time-consuming
samps.extend(samp)
U = set(samps)
else:
covered = set([t for r in self.rules for t in r.trans()])
U = self.sample_from_each_label(labels_samp, nsamp, covered,
mode)
print('nsamp (after):', len(U))
if len(U) == 0:
break
U_all.extend(U)
# Greedy
greed.update_univ(U)
r = greed.greedy_once()
# Termination criteria. Also check zero sampling above.
if self.enough(r):
# Include at least one rule per class, except default class.
labels_samp.remove(r.label)
print('remove label:', r.label)
else:
# Print quality vs. dispersion
q, d = obj(self.rules, lamb, sep=True)
qr, dr = obj(self.rules + [r], lamb, sep=True)
print('inc q vs. d: {}, {}'.format(qr - q, dr - d))
self.add(r)
if np.abs(recall(self.rules)[r.label] - 1.0) < 1e-8:
labels_samp.remove(r.label)
print('#{} '.format(len(self.rules)), end='')
printRules([r])
# Consecutive greedy over all sampels
if rerun:
greed.clear()
greed.update_univ(list(set(U_all)))
rules = greed.greedy(len(self.rules))
if obj(rules, lamb) > obj(self.rules, lamb):
print('Full greedy wins: {} > {}'.format(
obj(rules, lamb), obj(self.rules, lamb)))
self.reset(rules)
default = self.set_default()
print('default:', default)
self.build()
print('precision: ', precision(self).items())
print('recall (coverage): ', recall(self.rules).items())
print('ave disp: ', dispersion(self.rules, average=True))
print('##### END #####')