def get_random_transactions(transaction_number=500,
max_item_per_transaction=100, max_key_length=50,
key_alphabet=string.ascii_letters, universe_size=1000):
'''Generates a random list of `transaction_number` transactions containing
from 0 to `max_item_per_transaction` from a collection of
`universe_size`. Each key has a maximum length of `max_key_length` and
is computed from a sequence of characters specified by `key_alphabet`
(default is ascii letters).
If `key_alphabet` is None, range(universize_size) is used as the
alphabet and `max_key_length` is ignored.
'''
if key_alphabet is None:
words = list(range(universe_size))
else:
words = []
for _ in range(universe_size):
word = ''.join((random.choice(key_alphabet) for x in
range(random.randint(1, max_key_length))))
words.append(word)
transactions = []
for _ in range(transaction_number):
transaction = {word for word in random.sample(words, random.randint(0,
max_item_per_transaction))}
transactions.append(transaction)
return transactions
def get_random_transactions(transaction_number=500,
max_item_per_transaction=100,
max_key_length=50,
key_alphabet=string.ascii_letters,
universe_size=1000):
'''Generates a random list of `transaction_number` transactions containing
from 0 to `max_item_per_transaction` from a collection of
`universe_size`. Each key has a maximum length of `max_key_length` and
is computed from a sequence of characters specified by `key_alphabet`
(default is ascii letters).
If `key_alphabet` is None, range(universize_size) is used as the
alphabet and `max_key_length` is ignored.
'''
if key_alphabet is None:
words = list(range(universe_size))
else:
words = []
for _ in range(universe_size):
word = ''.join((random.choice(key_alphabet)
for x in range(random.randint(1, max_key_length))))
words.append(word)
transactions = []
for _ in range(transaction_number):
transaction = {
word
for word in random.sample(
words, random.randint(0, max_item_per_transaction))
}
transactions.append(transaction)
return transactions
def test_itemset_perf(perf_round=10, sparse=True, seed=None):
'''Non-scientifically tests the performance of three algorithms by running
`perf_round` rounds of FP-Growth, FP-Growth without pruning, Relim, and
SAM.
A random set of transactions is created (the same is obviously used
for all algorithms).
If `sparse` is False, the random transactions are more dense, i.e., some
elements appear in almost all transactions.
The `seed` parameter can be used to obtain the same sample across
multiple calls.
'''
random.seed(seed)
if sparse:
universe_size = 2000
transaction_number = 500
support = 10
else:
universe_size = 110
transaction_number = 75
support = 25
transactions = get_random_transactions(
transaction_number=transaction_number,
universe_size=universe_size,
key_alphabet=None)
print('Random transactions generated with seed {0}\n'.format(seed))
start = time()
for i in range(perf_round):
(n, report) = test_fpgrowth(False, transactions, support,
pruning=True)
print('Done round {0}'.format(i))
end = time()
print('FP-Growth (pruning on) took: {0}'.format(end - start))
print('Computed {0} frequent item sets.'.format(n))
start = time()
for i in range(perf_round):
(n, report) = test_fpgrowth(False, transactions, support,
pruning=False)
print('Done round {0}'.format(i))
end = time()
print('FP-Growth (pruning off) took: {0}'.format(end - start))
print('Computed {0} frequent item sets.'.format(n))
start = time()
for i in range(perf_round):
(n, report) = test_relim(False, transactions, support)
print('Done round {0}'.format(i))
end = time()
print('Relim took: {0}'.format(end - start))
print('Computed {0} frequent item sets.'.format(n))
start = time()
for i in range(perf_round):
(n, report) = test_sam(False, transactions, support)
print('Done round {0}'.format(i))
end = time()
print('Sam took: {0}'.format(end - start))
print('Computed {0} frequent item sets.'.format(n))
def test_itemset_perf(perf_round=10, sparse=True, seed=None):
'''Non-scientifically tests the performance of three algorithms by running
`perf_round` rounds of FP-Growth, FP-Growth without pruning, Relim, and
SAM.
A random set of transactions is created (the same is obviously used
for all algorithms).
If `sparse` is False, the random transactions are more dense, i.e., some
elements appear in almost all transactions.
The `seed` parameter can be used to obtain the same sample across
multiple calls.
'''
random.seed(seed)
if sparse:
universe_size = 2000
transaction_number = 500
support = 10
else:
universe_size = 110
transaction_number = 75
support = 25
transactions = get_random_transactions(
transaction_number=transaction_number,
universe_size=universe_size,
key_alphabet=None)
print('Random transactions generated with seed {0}\n'.format(seed))
start = time()
for i in range(perf_round):
(n, report) = test_fpgrowth(False, transactions, support, pruning=True)
print('Done round {0}'.format(i))
end = time()
print('FP-Growth (pruning on) took: {0}'.format(end - start))
print('Computed {0} frequent item sets.'.format(n))
start = time()
for i in range(perf_round):
(n, report) = test_fpgrowth(False,
transactions,
support,
pruning=False)
print('Done round {0}'.format(i))
end = time()
print('FP-Growth (pruning off) took: {0}'.format(end - start))
print('Computed {0} frequent item sets.'.format(n))
start = time()
for i in range(perf_round):
(n, report) = test_relim(False, transactions, support)
print('Done round {0}'.format(i))
end = time()
print('Relim took: {0}'.format(end - start))
print('Computed {0} frequent item sets.'.format(n))
start = time()
for i in range(perf_round):
(n, report) = test_sam(False, transactions, support)
print('Done round {0}'.format(i))
end = time()
print('Sam took: {0}'.format(end - start))
print('Computed {0} frequent item sets.'.format(n))