def run():
( train, undef, test ) = data.load_data_wrapper()
log( "Trainning set : %8d instances", ( len(train) ) )
log( "Testing set : %8d instances", ( len(test) ) )
txt = "Converting %s set to Instance objects"
log( txt, ("training") )
train = [ Instance( t[0], label_array=t[1] ) for t in train ]
log( txt, ("test") )
test = [ Instance( t[0], label=t[1] ) for t in test ]
instance = 1
start_compare = time.time()
""" Iterate through testing set """
test_subset = test[0:10]
for i in test_subset:
log("Instance %d", ( instance ))
""" Find the closest pair from training set
"""
closest_pair = train[0]
max_dist = cosine_dist( i, closest_pair )
for j in train[1:1000]:
dist = cosine_dist( i, j )
if( dist < max_dist ):
max_dist = dist
closest_pair = j
if( dist == 0 ):
break
i.predicted_label = closest_pair.label
log(">>> %d, actual : %s , predict : %s", ( instance, test[0].label, test[0].predicted_label) )
instance+=1
end_compare = time.time()
""" Compute confusion_matrix, accuracy and prediction and recall for each label
"""
log("----- Confusion Matrix -----")
matrix = confusion_matrix( test_subset )
log("%s", ( pandas.DataFrame( matrix ) ) )
log("Accuracy : %0.2f", ( accuracy(matrix) ) )
for i in range(NUM_LABEL):
log("Label %d : precision: %.2f \t recall: %.2f",
( i, precision( matrix, i ), recall( matrix, i ) )
)
log("----------------")
log("Time spent : %.0f sec", ( end_compare - start_compare ) )
def run():
parser = OptionParser()
parser.add_option("-k", "--dimension", dest="k", type="int", default=50 )
parser.add_option("-n", "--data", dest="n", type="int", default=20 )
(opt, args) = parser.parse_args()
(train, validation, test) = data_loader.load_data_wrapper();
train = train[:opt.n]
dim = len(train[0][0])
print( "k: %d" % ( opt.k ) )
print( "dimension: %d" % (dim))
print("---")
R = RandomProjection( opt.k, dim )
train = [ Sample( t[0], R.project(t[0]), t[1] ) for t in train ]
total = 0;
f = open( "k-"+str(opt.k)+".csv", 'wt')
try:
writer = csv.writer(f)
writer.writerow( ('Instance 1', 'Instance 2', 'Distortion') )
for i in range(len(train)):
a = train[i]
for j in range(i+1,len(train)):
if i == j:
continue
total+=1
b = train[j]
dist_k_dim = np.linalg.norm( a.low_features - b.low_features )
dist_d_dim = np.linalg.norm( a.features - b.features )
distortion = dist_k_dim / dist_d_dim
writer.writerow( (i+1, j+1, "%.4f" % ( distortion ) ))
finally:
f.close()
# total should be C(20,2)
print(total)
import numpy as np
import mnist_dataloader, time
# load data set
training_data, validation_data, test_data = mnist_dataloader.load_data_wrapper()
start_time = time.time()
# compute length of each instance in training_data
training_data_lengths = [np.linalg.norm(training_instance[0]) for training_instance in training_data]
# compute the length of each instance in test_data
test_data_lengths = [np.linalg.norm(test_instance[0]) for test_instance in test_data]
# for i in range(1, 100):
# print test_data_length[i]
# classify test_data
classified_results = []
for test_instance, test_instance_index in zip(test_data, range(len(test_data))):
# find the nearest training instance with cosine similarity
maximal_cosine_similarity = -1
maximal_cosine_similarity_index = 0
for training_instance, training_instance_index in zip(training_data, range(len(training_data))):
# compute the cosine similarity
# first, compute the inner product
inner_product = np.inner(test_instance[0].reshape(-1), training_instance[0].reshape(-1))
normalized_inner_product = inner_product / test_data_lengths[test_instance_index] / training_data_lengths[training_instance_index]
import numpy as np
import mnist_dataloader, time
# load data set
training_data, validation_data, test_data = mnist_dataloader.load_data_wrapper(
)
start_time = time.time()
# compute length of each instance in training_data
training_data_lengths = [
np.linalg.norm(training_instance[0]) for training_instance in training_data
]
# compute the length of each instance in test_data
test_data_lengths = [
np.linalg.norm(test_instance[0]) for test_instance in test_data
]
# for i in range(1, 100):
# print test_data_length[i]
# classify test_data
classified_results = []
for test_instance, test_instance_index in zip(test_data,
range(len(test_data))):
# find the nearest training instance with cosine similarity
maximal_cosine_similarity = -1
maximal_cosine_similarity_index = 0
def accuracy( confusion_matrix ):
total_instance = 0
correct_predict = 0
for i in range(NUM_LABEL):
total_instance = total_instance + sum( confusion_matrix[i] )
correct_predict = correct_predict + confusion_matrix[i][i]
return correct_predict*1.0 / total_instance
def log( format, data=() ):
text = format % data
print(text)
if __name__ == '__main__':
parser = OptionParser()
parser.add_option("-d","--dist_metric: 1.cosine 2.euclidean", dest="dist_type", type="string" ,default ="euclidean")
parser.add_option("-r","--reduce dimensions: Y(es) ,N(o)", dest="reduce_dims_flag", type="string", default="yes")
parser.add_option("-k", "--new_dimensions", dest="k", type="int", default=100 )
parser.add_option("-n", "--train_data_len", dest="n", type="int", default=1000)
parser.add_option("-t", "--test_data_len", dest="t", type="int", default=100)
(opt, args) = parser.parse_args()
print("distance_metric: "+opt.dist_type)
print("reduce_dims_flag: "+opt.reduce_dims_flag)
print("data lenght: "+str(opt.n))
print("new dinmesnions: "+str(opt.k))
(train, validation, test) = data_loader.load_data_wrapper(opt.n,opt.t)
runNNClassifier(train,test,opt.dist_type, opt.reduce_dims_flag.lower(), opt.k)