def __init__(self, dimensions=0, k_neighbors=5, use_kernel=False):
'''
Constructor for FaceRecognizer.
Args (optional):
k_rank (int): How many principal components to keep.
k_neighbors (int): How many neighbors to compare against in the
kNN classifier.
'''
self.pca_model = PCAModel(dimensions=dimensions, use_kernel=use_kernel)
self.knn_classifier = KNNClassifier(neighbors=k_neighbors)
self.instances = None
def kneighbors(tDataX, tDataY, mineData):
X_train, X_test, y_train, y_test = cross_validation.train_test_split(
tDataX, tDataY, test_size=0.4, random_state=0)
# X_train: The Attributes from the test data that will be used to train the model
# X_test: The Attributes from the test data that will be used to test the model
# y_train: The classification from the test data that will be used to train the model
# y_train: The classification from the test data that will be used to test the model
# A good rule of thumb is that k = the sqrt of n. Round to nearest int
n_neighbors = int(np.sqrt(len(X_train)))
weights = 'uniform'
# we create an instance of Neighbours Classifier and fit the data.
knn = nClass.KNN(n_neighbors, weights, X_train, y_train)
test_pred = knn.predictDataSet(X_test)
accuracy = knn.accuracy(X_test, y_test)
actual_pred = knn.predictDataSet(mineData)
resp = {
'Testset Class': y_test,
'Predicted Testset Class': test_pred,
'Classes': actual_pred
}
resp['mineAttrs'] = mineData
resp['accuracy'] = accuracy
return resp
def run(url, header, categories=None, Nan="?", k=3):
data = read(url, header)
#print data
if categories is not None:
for key in categories:
conversion = categorize(data[key], categories[key], Nan=Nan)
#print conversion.keys()
data[key] = data[key].map(conversion)
#for x in xrange(len(data[key])):
# #print data[key][x]
# if data[key][x] in conversion.keys():
# print "converting..."
# data[key][x] = conversion[data[key][x]]
else:
for head in header:
if numpy.issubdtype(data[head].dtype, numpy.number):
conversion = categorize(data[head], build_options_list(data[head], Nan))
data[head] = data[head].map(conversion)
print data[header[13]][14]
#data.replace(Nan, "0")
#for key in header:
# #print data[key]
# for y in xrange(len(data[key])):
# if data[key][y] == "?":
# print "Guessing..."
# data[key][y] = 0.0
print data
print data[header[13]][14]
data.apply(pandas.to_numeric)
print data[header[13]][14]
data = data.as_matrix()
print data[14][13]
print data
print "data tested with K = ", k, ": ", KNNClassifier.test_classifier(data[:, range(13)], data[:, 14], k)
def __init__(self, dimensions=0, k_neighbors=5, use_kernel=False):
'''
Constructor for FaceRecognizer.
Args (optional):
k_rank (int): How many principal components to keep.
k_neighbors (int): How many neighbors to compare against in the
kNN classifier.
'''
self.pca_model = PCAModel(dimensions=dimensions, use_kernel=use_kernel)
self.knn_classifier = KNNClassifier(neighbors=k_neighbors)
self.instances = None
def main():
df_beijing = pd.read_csv(BEIJING_PATH)
df_shenyang = pd.read_csv(SHENYANG_PATH)
X, Y = prepare_data_and_labels(df_beijing, df_shenyang)
clf = KNNClassifier(K=5)
validate(clf, X, Y)
print("GUANGZHOU")
df_guangzhou = pd.read_csv(GUANGZHOU_PATH)
X_test, Y_test = prepare_data_and_labels(df_guangzhou)
test(clf, X_test, Y_test)
print("SHANGHAI")
df_shanghai = pd.read_csv(SHANGHAI_PATH)
X_test, Y_test = prepare_data_and_labels(df_shanghai)
test(clf, X_test, Y_test)
def classify(folder, main_folder,num_class, num_feature,F, K):
#for folder in address:
feature_num_class = []
with open(main_folder+"/feature_num_of_classes.csv",'rU') as f:
reader = csv.reader(f)
for row in reader:
feature_num_class.append(row)
f.close()
binary_class_labels = []
with open(main_folder+"/binary_class_labels.csv",'rU') as f:
reader = csv.reader(f)
for row in reader:
binary_class_labels.append(row)
f.close()
feature_vectors = []
with open(main_folder+"/feature_vectors_trim.csv",'rU') as f:
reader = csv.reader(f)
for row in reader:
feature_vectors.append(row)
f.close()
result = open(folder+"/knn_results.txt","w")
class_frequency = []
feature_weights = []
training_article_nums = []
test_article_nums = []
sample_training_article_nums = []
sample_test_article_nums = []
sample_training_vectors = []
sample_training_labels = []
sample_test_vectors = []
sample_test_labels = []
with open(folder+"/training_article_numbers.csv", 'rb') as f:
reader=csv.reader(f)
for row in reader:
training_article_nums.append(row)
f.close()
with open(folder+"/test_article_numbers.csv", 'rb') as f:
reader=csv.reader(f)
for row in reader:
test_article_nums.append(row)
f.close()
with open(main_folder+"/class_frequency.csv", 'rb') as f:
reader = csv.reader(f)
for row in reader:
class_frequency.append(row)
f.close()
with open(main_folder+"/weight_feature_class.csv", 'rU') as f:
reader = csv.reader(f)
for row in reader:
feature_weights.append(row)
f.close()
if False: from random import shuffle; X=list(training_article_nums[0]); shuffle(training_article_nums[0])
for k in xrange(F):
#sample_training = [x for i, x in enumerate(article_number) if i % K != k]
sample_training = [x for i, x in enumerate(training_article_nums[0]) if i % F == k]
with open(folder+"/sample_training_article_num.csv", 'wb') as f:
writer=csv.writer(f)
writer.writerow(sample_training)
f.close()
if False: from random import shuffle; X=list(test_article_nums[0]); shuffle(test_article_nums[0])
for k in xrange(F):
#sample_training = [x for i, x in enumerate(article_number) if i % K != k]
sample_test = [x for i, x in enumerate(test_article_nums[0]) if i % F == k]
with open(folder+"/sample_test_article_num.csv", 'wb') as f:
writer=csv.writer(f)
writer.writerow(sample_test)
f.close()
with open(folder+"/sample_training_article_num.csv", 'wb') as f:
writer=csv.writer(f)
writer.writerow(sample_training)
f.close()
with open(folder+"/sample_training_vectors.csv", 'wb') as f:
writer=csv.writer(f)
for item in sample_training:
sample_training_vectors.append(feature_vectors[int(item)])
writer.writerow(feature_vectors[int(item)])
f.close()
with open(folder+"/sample_training_labels.csv", 'wb') as f:
writer=csv.writer(f)
for item in sample_training:
sample_training_labels.append(binary_class_labels[int(item)])
writer.writerow(binary_class_labels[int(item)])
f.close()
with open(folder+"/sample_test_vectors.csv", 'wb') as f:
writer=csv.writer(f)
for item in sample_test:
sample_test_vectors.append(feature_vectors[int(item)])
writer.writerow(feature_vectors[int(item)])
f.close()
with open(folder+"/sample_test_labels.csv", 'wb') as f:
writer=csv.writer(f)
for item in sample_test:
sample_test_labels.append(binary_class_labels[int(item)])
writer.writerow(binary_class_labels[int(item)])
f.close()
modeling_time = 0
classifying_time=0
inversed_classified_labels = []
for i in range(0,num_class):
temp = []
feature_num = feature_num_class[i]
the_class_frequency = class_frequency[0][i]
start_time = time.time()
class_column = []
for item in sample_training_labels :
if item[0]=='':
break
class_column.append(item[i])
knn=KNNClassifier.train(sample_training_vectors, class_column, K, typecode=None)
modeling_time+=(time.time()-start_time)
#print "modeling time = ", modeling_time, "s"
#print "classifier number = ",i+1
for article in sample_test_vectors:
start_time = time.time()
if article[0]=='':
break
label = KNNClassifier.classify(knn, article,feature_num, the_class_frequency,None , feature_weights[0])
classifying_time += (time.time()-start_time)
#print "classifying time = ", classifying_time,"s"
#print "class number = ",i," label = ", label
temp.append(label)
print( i+1,"out of ",num_class," has been classified")
inversed_classified_labels.append(temp)
print("modeling time = ", modeling_time,"s", file = result)
print("number of knn classifiers = ", num_class, file = result)
print("classifying time = ", classifying_time, "s", file = result)
#inverse-the inversed label matrix
num_article = len(inversed_classified_labels[0])
print("calssified article number = ", num_article, file = result)
with open(folder+"/knn_classified_sample_labels.csv", 'wb') as f:
writer = csv.writer(f)
for i in range(0,num_article):
temp = []
for j in range(0,num_class):
temp.append(inversed_classified_labels[j][i])
writer.writerow(temp)
f.close()
calculate_accuracy()
nrows=2,
ncols=5,
sharex=True,
sharey=True,
)
ax = ax.flatten()
for i in range(10):
img = X_train[y_train == i][0]
ax[i].imshow(img, cmap='Greys', interpolation='nearest')
ax[0].set_xticks([])
ax[0].set_yticks([])
plt.tight_layout()
plt.show()
X_train = v.NormalVectorize(X_train)
X_test = v.NormalVectorize(X_test)
"""Average Classifier"""
AC = AverageClassifier.Train(X_train, y_train)
print("Average Classifier Accurary:",
AverageClassifier.Test(X_test, y_test, AC) * 100, "%")
"""KNN Classifier"""
print("KNN Predict Classifier:")
print("KNN")
KNNClassifier.Test(X_test, y_test, X_train, y_train)
print("KNN with Lib")
KNN = KNNClassifier.KNNbyLib(X_train, y_train)
KNNClassifier.KNNByLibTest(X_test, y_test, KNN)
"""ANN Classifier"""
ANN = ANNClassifier.Train(X_train, y_train)
print("ANN Classifier Accurary:")
ANNClassifier.Test(X_test, y_test, ANN)
import numpy as np
import KNNClassifier
import matplotlib.pyplot as plt
if __name__ == '__main__':
image_size = 28 # width and length
no_of_different_labels = 10 # i.e. 0, 1, 2, 3, ..., 9
image_pixels = image_size * image_size
data_path = "/mnist/"
train_data = np.loadtxt(data_path + "mnist_train.csv", delimiter=",")
test_data = np.loadtxt(data_path + "mnist_test.csv", delimiter=",")
fac = 0.99 / 255
train_imgs = np.asfarray(train_data[:, 1:]) * fac + 0.01
test_imgs = np.asfarray(test_data[:, 1:]) * fac + 0.01
train_labels = np.asfarray(train_data[:, :1])
test_labels = np.asfarray(test_data[:, :1])
knn = KNNClassifier(distance='Euclidean', K=5)
knn.fit(train_imgs, train_labels)
results = knn.predict(test_imgs)
#
# for i in range(10):
# img = train_imgs[i].reshape((28,28))
# plt.imshow(img, cmap="Greys")
# plt.show()
from sklearn.metrics import confusion_matrix
confusion_matrix(test_labels, results)
###########################################
## KNN test kodları
from sklearn.datasets import load_iris
from sklearn.utils import shuffle
iris_X, iris_y = load_iris(return_X_y=True)
iris_X, iris_y = shuffle(iris_X, iris_y)
X_train = iris_X[:-30]
X_test = iris_X[-30:]
y_train = iris_y[:-30]
y_test = iris_y[-30:]
from KNNClassifier import *
knn = KNNClassifier("eucledean", 10)
knn.buildModel(X_train, y_train)
knn.evaluateModel(X_test, y_test)
knn.showLabel(X_test[5], load_iris())
#########################################
## Naive bayes test kodları
from sklearn.datasets import load_iris
from sklearn.utils import shuffle
iris_X, iris_y = load_iris(return_X_y=True)
iris_X, iris_y = shuffle(iris_X, iris_y)
X_train = iris_X[:-30]
X_test = iris_X[-30:]
y_train = iris_y[:-30]
class FaceRecognizer:
def __init__(self, dimensions=0, k_neighbors=5, use_kernel=False):
'''
Constructor for FaceRecognizer.
Args (optional):
k_rank (int): How many principal components to keep.
k_neighbors (int): How many neighbors to compare against in the
kNN classifier.
'''
self.pca_model = PCAModel(dimensions=dimensions, use_kernel=use_kernel)
self.knn_classifier = KNNClassifier(neighbors=k_neighbors)
self.instances = None
def train(self, instances):
'''
Trains the recognizer with a set of faces.
Args:
instances (list<tuple<int, numpy.ndarray>>): List of label/face
data pairs.
'''
self.instances = instances
# Stack all of the faces together
faces_list = list()
for instance in instances:
faces_list.append(instance[1])
faces = np.vstack(faces_list).T
# Learn principal components
self.pca_model.fit(faces)
# Add each class to the kNN classifier
for instance in instances:
label = instance[0]
face = instance[1]
t_face = self.pca_model.transform(face)
self.knn_classifier.add_sample(label, t_face)
def fit_knn(self):
'''
Fits the kNN classifier with the current instances.
'''
if self.instances is None:
raise RuntimeError('FaceRecognizer has no instances')
self.knn_classifier.reset()
for instance in self.instances:
label = instance[0]
face = instance[1]
t_face = self.pca_model.transform(face)
self.knn_classifier.add_sample(label, t_face)
def classify(self, face):
'''
Classifies a given face from the trained set.
Args:
face (numpy.ndarray): The face to classify.
Returns:
int, the class the face best belongs to.
'''
t_face = self.pca_model.transform(face)
return self.knn_classifier.classify(t_face)
def set_dimensions(self, dimensions):
'''
Sets the number of dimensions to use from PCA.
Args:
dimensions (int): The new number of dimensions.
'''
self.pca_model.dimensions = dimensions
if self.instances is not None:
self.fit_knn()
def set_k_neighbors(self, k):
'''
Sets k for kNN classifier.
Args:
k (int): The new k for the classifier.
'''
self.knn_classifier.neighbors = k
def set_kernel_variance(self, variance):
'''
Sets the variance for the RBF kernel and retrains.
Args:
variance (float): The new variance.
'''
self.pca_model.variance = variance
if self.instances is not None and self.pca_model.use_kernel:
self.train(self.instances)
class FaceRecognizer:
def __init__(self, dimensions=0, k_neighbors=5, use_kernel=False):
'''
Constructor for FaceRecognizer.
Args (optional):
k_rank (int): How many principal components to keep.
k_neighbors (int): How many neighbors to compare against in the
kNN classifier.
'''
self.pca_model = PCAModel(dimensions=dimensions, use_kernel=use_kernel)
self.knn_classifier = KNNClassifier(neighbors=k_neighbors)
self.instances = None
def train(self, instances):
'''
Trains the recognizer with a set of faces.
Args:
instances (list<tuple<int, numpy.ndarray>>): List of label/face
data pairs.
'''
self.instances = instances
# Stack all of the faces together
faces_list = list()
for instance in instances:
faces_list.append(instance[1])
faces = np.vstack(faces_list).T
# Learn principal components
self.pca_model.fit(faces)
# Add each class to the kNN classifier
for instance in instances:
label = instance[0]
face = instance[1]
t_face = self.pca_model.transform(face)
self.knn_classifier.add_sample(label, t_face)
def fit_knn(self):
'''
Fits the kNN classifier with the current instances.
'''
if self.instances is None:
raise RuntimeError('FaceRecognizer has no instances')
self.knn_classifier.reset()
for instance in self.instances:
label = instance[0]
face = instance[1]
t_face = self.pca_model.transform(face)
self.knn_classifier.add_sample(label, t_face)
def classify(self, face):
'''
Classifies a given face from the trained set.
Args:
face (numpy.ndarray): The face to classify.
Returns:
int, the class the face best belongs to.
'''
t_face = self.pca_model.transform(face)
return self.knn_classifier.classify(t_face)
def set_dimensions(self, dimensions):
'''
Sets the number of dimensions to use from PCA.
Args:
dimensions (int): The new number of dimensions.
'''
self.pca_model.dimensions = dimensions
if self.instances is not None:
self.fit_knn()
def set_k_neighbors(self, k):
'''
Sets k for kNN classifier.
Args:
k (int): The new k for the classifier.
'''
self.knn_classifier.neighbors = k
def set_kernel_variance(self, variance):
'''
Sets the variance for the RBF kernel and retrains.
Args:
variance (float): The new variance.
'''
self.pca_model.variance = variance
if self.instances is not None and self.pca_model.use_kernel:
self.train(self.instances)
