Upload New File

ReadMe.md

+# Image classification
+
+Antonin DELORME - TD 1 sur le cours Deep Learning 
+
+
+## Pepare the CIFAR dataset 
+
+### 2.
+```rb
+def read_cifar_batch(path):
+    """ path = "data\cifar-10-batches-py\data_batch_1" 
+    par exemple """
+    with open(path, 'rb') as fo:
+        dict = pickle.load(fo, encoding='bytes')
+    
+    labels=list(dict.items())[1][1] #labels[i] est le label de l'ième image
+    data=list(dict.items())[2][1] #data[i] sont les 3072 pixel de l'image i
+    return (labels,data)
+
+
+
+```
+
+### 3.
+```rb
+def read_cifar(path):
+    """ path="data\cifar-10-batches-py" par exemple """
+    (labels,data)=read_cifar_batch(path+"\\test_batch")
+    
+    for i in range(1,6):
+        data=np.concatenate((data,read_cifar_batch(path+"\\data_batch_"+str(i))[1]),axis=0)
+        labels=labels+read_cifar_batch(path+"\\data_batch_"+str(i))[0]
+    return (labels,data)
+```
+### 4. 
+
+```rb
+def split_dataset(labels,data,split):
+    split=round(split*len(labels))
+    test=[]
+    while len(test) != split:
+        Nb=rd.randint(0,len(labels)-1)
+        if Nb not in test :
+            test.append(Nb)
+    train=[i for i in range(len(labels)) if i not in test]
+    
+    data_train=data[train]
+    data_test=data[test]
+    labels_test=[]
+    labels_train=[]
+    for i in test:
+        labels_test.append(labels[i])
+    for j in train:
+        labels_train.append(labels[j])
+        
+    return(data_train,labels_train,data_test,labels_test)
+            
+
+
+```
+
+
+## KNN
+### 1.
+```rb
+
+
+def distance_matrix(a,b):
+    at=np.transpose(a)
+    bt=np.transpose(b)
+    som_carr_a=np.sum(np.square(a), axis=1, keepdims=True)
+    som_carr_b=np.sum(np.square(b), axis=1, keepdims=True)
+    
+    prod=np.dot(a,bt)
+    
+    return ( np.sqrt(som_carr_a + np.transpose(som_carr_b) - 2 * prod) )
+
+```
+
+### 2.
+
+```rb
+
+def knn_predict(dists,labels_train,k):
+    predict=[]
+    for i in range(len(dists)):
+        Glob_dist=[]
+        Glob_min=[]
+        for j in range(len(dists[i])):
+            Glob_dist.append(dists[i][j])
+        for p in range(k):
+            m = min(Glob_dist)
+            index = Glob_dist.index(m)
+            Glob_min.append(labels_train[index])
+            del(Glob_dist[index])
+        
+        Temp=np.bincount(Glob_min)
+        predict.append(list(Temp).index(max(Temp)))
+            
+            
+    return predict
+
+```
+### 3.
+
+```rb
+def evaluate_knn(data_train,data_test,labels_train,labels_test,k,dist):
+    pred=knn_predict(dist, labels_train, k)
+    tot=0
+    bon=0
+    for i in range(len(labels_test)):
+        if labels_test[i]==pred[i]:
+            bon+=1
+        tot+=1
+    
+    print("Accuracy :",bon/tot)
+    return bon/tot
+    
+
+
+
+```
+
+### 4.
+
+
+![Alt text](<Knn res.png>)
\ No newline at end of file