多標記KNN演算法實現（Python3 6）

對於乙個新例項，取其最近的k個例項，然後得到由k個例項組成的標籤集合，最後通過先驗概率與最大後驗概率來確定新例項的標籤集合。詳細的內容參看周志華老師與張敏靈老師的《多標記學習》

演算法實現

資料預處理

mlknndemo.py

#load file

data = sio.loadmat('scene.mat')
train_bags = data['train_bags']
test_bags = data['test_bags']
train_targets = data['train_targets']
test_targets = data['test_targets']
#train_bags:9*15 to 1*135
trainbagline = len(train_bags)
train_data = 
for i in range(trainbagline):
linshi = train_bags[i,0].flatten().tolist()
train_data=np.array(train_data)
#test_bags:9*15 to 1*135
testbagline = len(test_bags)
test_data = 
for i in range(testbagline):
linshi = test_bags[i,0].flatten().tolist()
test_data=np.array(test_data)
#const;num is k's value
num = 10
smooth = 1
#training;
prior,priorn,cond,condn = mlknntrain.trainclass(train_data,train_targets,num,smooth)
#testing
outputs,prelabels = mlknntest.testclass(train_data,train_targets,test_data,test_targets,num,prior,priorn,cond,condn)

mlknntrain.py

#train 

def trainclass(train_data,train_targets,num,smooth):
#get size of matrix
num_class,num_training = np.mat(train_targets).shape 
dist_matrix = np.diagflat(ones((1,num_training))*sys.maxsize)
#cumputing distance
for i in range((num_training-1)):
vector1 = train_data[i,:]
for j in range((i+1),(num_training)):
vector2 = train_data[j,:]
dist_matrix[i,j] = sum((vector1-vector2)**2)**0.5
dist_matrix[j,i] = dist_matrix[i,j]
#prior and priorn
prior = zeros((num_class,1))
priorn = zeros((num_class,1))
for i in range(num_class):
tempci = sum((train_targets[i,:]==ones((1,num_training))))
prior[i,0] = (tempci+1)/(smooth*2+num_training)
priorn[i,0] = 1-prior[i,0]
#cond and condn
#sort by distance and get index
#find neighbors
dismatindex = argsort(dist_matrix)
tempci = zeros((num_class,num+1))
tempnci = zeros((num_class,num+1))
for i in range(num_training):
temp = zeros((1,num_class))
neighborlabels = 
for j in range(num):
neighborlabels = np.mat(neighborlabels)
neighborlabels = np.transpose(neighborlabels)
for j in range(num_class):
temp[0,j] = sum((neighborlabels[j,:] == ones((1,num))))
for j in range(num_class):
t = int((temp[0,j]))
if(train_targets[j,i] == 1):
tempci[j,t]=tempci[j,t]+1
else:
tempnci[j,t] = tempnci[j,t]+1
#get 5*11 matrix
cond = zeros((num_class,num+1))
condn = zeros((num_class,num+1))
for i in range(num_class):
temp1 = sum((tempci[i,:]))
temp2 = sum((tempnci[i,:]))
for j in range(num+1):
cond[i,j] = (smooth+tempci[i,j])/(smooth*(num+1)+temp1)
condn[i,j] = (smooth+tempnci[i,j])/(smooth*(num+1)+temp2)
return prior,priorn,cond,condn

測試mlknntest.py

def testclass(train_data,train_targets,test_data,test_targets,num,prior,priorn,cond,condn):

num_class,num_training = np.mat(train_targets).shape
num_class,num_testing = np.mat(test_targets).shape
#init matrix about distance
distmatrix = zeros((num_testing,num_training))
for i in range(num_testing):
vector1 = test_data[i,:]
for j in range(num_training):
vector2 = train_data[j,:]
distmatrix[i,j] = sum((vector1-vector2)**2)**0.5
#sort by distance and get index
#find neighbors
dismatindex = argsort(distmatrix)
#computing outputs
outputs = zeros((num_class,num_testing))
for i in range(num_testing):
temp = zeros((1,num_class))
neighborlabels = 
for j in range(num):
neighborlabels = np.mat(neighborlabels)
#transposition
neighborlabels = np.transpose(neighborlabels)
for j in range(num_class):
temp[0,j] = sum((neighborlabels[j,:] == ones((1,num))))
for j in range(num_class):
t = int((temp[0,j]))
prob_in=prior[j]*cond[j,t]
prob_out=priorn[j]*condn[j,t]
if((prob_in+prob_out)==0):
outputs[j,i]=prior[j]
else:
outputs[j,i]=prob_in/[prob_in+prob_out]
#evaluation
prelabels=zeros((num_class,num_testing))
for i in range(num_testing):
for j in range(num_class):
if(outputs[j,i]>=0.5):                             #閾值為0.5
prelabels[j,i]=1
else:
prelabels[j,i]=-1
return outputs,prelabels

knn演算法實現

knn演算法 自己實現 鳶尾花資料集 一.題目描述 題目 自己實現knn演算法 用鳶尾花資料集 knn演算法描述 在訓練集中資料和標籤已知的情況下，輸入測試資料，將測試資料的特徵與訓練集中對應的特徵進行相互比較，找到訓練集中與之最為相似的前k個資料，則該測試資料對應的類別就是k個資料 現次數最多的那...

KNN演算法實現

knn k 近鄰 knn,k nearestneighbor 演算法是一種基本分類與回歸方法，我們這裡只討論分類問題中的 k 近鄰演算法。k 近鄰演算法的輸入為例項的特徵向量，對應於特徵空間的點 輸出為例項的類別，可以取多類。k 鄰演算法假設給定乙個訓練資料集，其中的例項類別已定。分類時，對新的例項...

python實現演算法 kmp演算法python實現

kmp演算法python實現 kmp演算法 kmp演算法用於字串的模式匹配，也就是找到模式字串在目標字串的第一次出現的位置 比如abababc 那麼bab在其位置1處，bc在其位置5處 我們首先想到的最簡單的辦法就是蠻力的乙個字元乙個字元的匹配，但那樣的時間複雜度會是o m n kmp演算法保證了時...