def run_for_k(k, n):
print(f"running kNN for {n} images with k={k}")
correct_predictions = 0
for i in range(len(test)):
prediction = knn(train[:n], train_labels, test[i], k=k)
if prediction == test_labels[i]:
correct_predictions += 1
print(f"number of correct predictions for (k={k}, n={n}): {correct_predictions}")
correct_predictions_rate = float(correct_predictions) / float(len(test))
print(f"correct predication percentage for (k={k},n={n}): {correct_predictions_rate}")
return correct_predictions_rate
from nn import knn, cosine_distance, gaussian_distance, polyd2_distance
if __name__=="__main__":
# Load data using specialized script
train_dataset = load_mnist(path="../data/mnist/", dataset="training")
test_dataset = load_mnist(path="../data/mnist/", dataset="testing")
# Take a fraction of the data to speed computation
train_images, train_labels = sample(train_dataset, 5000)
test_images, test_labels = sample(test_dataset, 1000)
# Get the bounds of the haar rectangles
bounds = genbounds(28, 28, 100)
# Create data, using same rectangles for training and testing
train_data = genfeatures(train_images, bounds)
test_data = genfeatures(test_images, bounds)
# Normalize the data
zmscaler = preprocessing.StandardScaler()
train_data = zmscaler.fit_transform(train_data)
test_data = zmscaler.transform(test_data)
# Run knn
for d in [cosine_distance, gaussian_distance, polyd2_distance]:
for k in [1,3,7]:
H = knn(train_data, test_data, train_labels, d=d, k=k)
c = np.sum(test_labels.ravel()==H)
print "k=%d:" % k, float(c)/float(len(test_labels))
def pubmed():
train = load_data("E:\\pubmed.csv")
m, n = train.shape
labels = load_data("E:\\pubmed_label.csv")
train_copy = train
list_k_fold = man_split(train_copy, labels, 5)
acc_bayes = 0
acc_funbayes = 0
acc_knn = 0
acc_funknn = 0
#print("done")
micro_bayes = 0
micro_funbayes = 0
micro_knn = 0
micro_funknn = 0
macro_bayes = 0
macro_funbayes = 0
macro_knn = 0
macro_funknn = 0
for k1 in range(5):
#print(k1)
#print("k1=",end=' ')
#print(k1)
test_set = []
training_set = []
training_label = []
test_label = []
prior = {}
label1 = []
#print(list_k_fold)
for i in range(5):
if i == k1:
label1.extend(list_k_fold[i])
#print("done")
for i2 in range(len(labels)):
if i2 in label1:
test_set.append(train_copy[i2])
test_label.append(labels[i2])
else:
training_set.append(train_copy[i2])
#print(trainset)
training_label.append(labels[i2])
#print("done")
dict_info = {}
dict_info = form_dict(training_set, training_label)
mean_dict = bay.find_mean(dict_info)
std_dict = bay.find_std(dict_info, mean_dict)
prior = bay.find_priors(dict_info)
_, predictions0 = bay.fun_bayes(training_set, test_set, training_label,
test_label)
acc_funbayes += _
micro_funbayes += f1_score(test_label, predictions0, average='micro')
macro_funbayes += f1_score(test_label, predictions0, average='macro')
#print(macro_funbayes)
#print("a")
_, predictions1 = nn.fun_knn(training_set, test_set, training_label,
test_label)
acc_funknn += _
micro_funknn += f1_score(test_label, predictions1, average='micro')
macro_funknn += f1_score(test_label, predictions1, average='macro')
#print("b")
_, predictions2 = nn.knn(training_set, training_label, test_set,
test_label)
acc_knn += _
micro_knn += f1_score(test_label, predictions2, average='micro')
macro_knn += f1_score(test_label, predictions2, average='macro')
_, predictions3 = bay.bayes(mean_dict, std_dict, test_set, test_label,
prior)
acc_bayes += _
micro_bayes += f1_score(test_label, predictions3, average='micro')
macro_bayes += f1_score(test_label, predictions3, average='macro')
#print("c")
file1.write("Test Accuracy on pubmed using inbuilt bayes ::" +
str(acc_funbayes / 5) + "\n")
file1.write("Test Accuracy on pubmed using inbuilt knn ::" +
str(acc_funknn / 5) + "\n")
file.write("Test Accuracy on pubmed using my bayes ::" +
str(acc_bayes / 5) + "\n")
file.write("Test Accuracy on pubmed using my knn ::" + str(acc_knn / 5) +
"\n \n")
file1.write("Test Macro F1 Score on pubmed using inbuilt bayes ::" +
str(macro_funbayes / 5) + "\n")
file1.write("Test Macro F1 Score on pubmed using inbuilt knn ::" +
str(macro_funknn / 5) + "\n")
file.write("Test Macro F1 Score on pubmed using my bayes ::" +
str(macro_bayes / 5) + "\n")
file.write("Test Macro F1 Score on pubmed using my knn ::" +
str(macro_knn / 5) + "\n \n")
file1.write("Test Micro F1 Score on pubmed using inbuilt bayes ::" +
str(micro_funbayes / 5) + "\n")
file1.write("Test Micro F1 Score on pubmed using inbuilt knn ::" +
str(micro_funknn / 5) + "\n")
file.write("Test Micro F1 Score on pubmed using my bayes ::" +
str(micro_bayes / 5) + "\n")
file.write("Test Micro F1 Score on pubmed using my knn ::" +
str(micro_knn / 5) + "\n \n")
training_label, test_label)
acc_funbayes += _
micro_funbayes += f1_score(test_label,
predictions0,
average='micro')
macro_funbayes += f1_score(test_label,
predictions0,
average='macro')
_, predictions1 = nn.fun_knn(training_set, test_set,
training_label, test_label)
acc_funknn += _
micro_funknn += f1_score(test_label, predictions1, average='micro')
macro_funknn += f1_score(test_label, predictions1, average='macro')
_, predictions2 = nn.knn(training_set, training_label, test_set,
test_label)
acc_knn += _
micro_knn += f1_score(test_label, predictions2, average='micro')
macro_knn += f1_score(test_label, predictions2, average='macro')
_, predictions3 = bay.bayes(mean_dict, std_dict, test_set,
test_label, prior)
acc_bayes += _
micro_bayes += f1_score(test_label, predictions3, average='micro')
macro_bayes += f1_score(test_label, predictions3, average='macro')
file1.write(
"Test Accuracy on dolphin using inbuilt bayes and d =" +
str(d) + "::" + str(acc_funbayes) + "\n")
file1.write("Test Accuracy on dolphin using inbuilt knn and d =" +
str(d) + "::" + str(acc_funknn) + "\n")
kernel = POLY
elif args.k == "l":
kernel = LINEAR
elif args.k == "r":
kernel = RBF
data_file = "../data/spambase/spambase.data"
dmat = []
f = open(data_file, "r")
for line in f:
x = line.split(',')
x = [float(e) for e in x]
dmat.append(x)
data = np.array(dmat)
# k-folds
k = 10
kfolder = KFolder(data, k, standard=True, shuffle=False)
for i in range(1):
print "Fold:", i+1
# Get data and labels at fold k
X,Y = kfolder.training(i)
# Get the testing data
Xi,Yi = kfolder.testing(i)
# Run knn
for j in [1,2,3]:
H = knn(X, Xi, Y, k=j)
c = np.sum(Yi.ravel()==H)
print "k=%d:" % j, float(c)/float(len(Yi))