for i in range(test_length):
#if rank_result[i] == test_label[i]:
if not (rank_result[i] ^ test_label[i]):
num_hits += 1
acc_list.append(float(num_hits) / float(test_length))
#Bloom Wisard
btuple_list = [2, 4, 8, 14, 16, 18, 20, 22, 24, 26, 28, 30, 32, 40, 56]
bacc_list = []
#capacity = len(train_bin)
capacity = 50
print capacity
for t in btuple_list:
bwisard = wnn.BloomWisard(entry_size, t, num_classes, capacity)
bwisard.train(train_bin, train_label)
rank_result = bwisard.rank(test_bin)
num_hits = 0
for i in range(test_length):
#if rank_result[i] == test_label[i]:
if not (rank_result[i] ^ test_label[i]):
num_hits += 1
bacc_list.append(float(num_hits) / float(test_length))
print "Tuples=", tuple_list
print "Wisard Accuracy=", acc_list
print "Tuples=", btuple_list
b_stats = []
b_training_time = []
b_testing_time = []
b_acc = []
b_error = []
for e in range(len(errors)):
btraining_time = []
btesting_time = []
bacc_list = []
for r in range(num_runs):
bwisard = wnn.BloomWisard(entry_size,
tuple_bit,
num_classes,
capacity,
error=errors[e])
#Training
start = timer()
bwisard.train(train_bin, train_label)
btraining_time.append(timer() - start)
#Testing
start = timer()
rank_result = bwisard.rank(test_bin)
btesting_time.append(timer() - start)
#Accuracy
num_hits = 0
wisard.train(folds_train_bin[f], folds_train_label[f])
rank_result = wisard.rank(folds_test_bin[f])
num_hits = 0
for i in range(test_length):
if rank_result[i] == folds_test_label[f][i]:
num_hits += 1
acc_list[j] += (float(num_hits)/float(test_length))
j += 1
#Bloom Wisard
j = 0
for t in btuple_list:
bwisard = wnn.BloomWisard(entry_size, t, num_classes, len(folds_train_label[f]))
bwisard.train(folds_train_bin[f], folds_train_label[f])
rank_result = bwisard.rank(folds_test_bin[f])
num_hits = 0
for i in range(test_length):
if rank_result[i] == folds_test_label[f][i]:
num_hits += 1
bacc_list[j] += (float(num_hits)/float(test_length))
j += 1
for i in range(len(tuple_list)):
bin_data = bin_data.reshape(entry_size, )
test_bin.append(bin_data)
train_bin = np.asarray(train_bin)
test_bin = np.asarray(test_bin)
#Wisard
num_classes = 10
test_length = len(test_label)
#Bloom Wisard
btuple_list = [2, 4, 7, 8, 14, 16, 28, 49, 56]
bacc_list = []
#for t in btuple_list:
bwisard = wnn.BloomWisard(entry_size, 28, num_classes, 50000, error=0.5)
bwisard.train(train_bin, train_label)
rank_result = bwisard.rank(test_bin)
bwisard_stats = bwisard.stats()
print bwisard_stats
num_hits = 0
for i in range(test_length):
if rank_result[i] == test_label[i]:
num_hits += 1
bacc_list.append(float(num_hits) / float(test_length))
#print "Tuples=", btuple_list
print "BloomWisard Accuracy=", bacc_list
rank_result = wisard.rank(test_bin)
num_hits = 0
for i in range(test_length):
if rank_result[i] == test_label[i]:
num_hits += 1
acc_list.append(float(num_hits) / float(test_length))
#Bloom Wisard
btuple_list = [2, 4, 8, 14, 16, 18, 20, 22, 24, 26, 28, 30, 32, 40, 56]
bacc_list = []
for t in btuple_list:
bwisard = wnn.BloomWisard(entry_size, t, num_classes, len(train_bin))
bwisard.train(train_bin, train_label)
rank_result = bwisard.rank(test_bin)
num_hits = 0
for i in range(test_length):
if rank_result[i] == test_label[i]:
num_hits += 1
bacc_list.append(float(num_hits) / float(test_length))
print "Tuples=", tuple_list
print "Wisard Accuracy=", acc_list
print "Tuples=", btuple_list
print "BloomWisard Accuracy=", bacc_list
#Wisard print "Wisard" wisard2 = wnn.Wisard(20, 4, 2) c = [a, b] #t = np.ndarray(shape=(2, 20), buffer=np.array(c, dtype=bool), dtype=bool) wisard2.train(c, [0, 1]) #print t print wisard2.rank(c) wisard2.info() #BloomWisard print "Bloom Wisard" bwisard = wnn.BloomWisard(20, 4, 2, 1000) bwisard.train(c, [0, 1]) print bwisard.rank(c) bwisard.info() print "Dict Wisard" wisard = wnn.DictWisard(20, 4, 2) wisard.train(c, [0, 1]) #print t print wisard.rank(c) wisard.info() print "Done"
