From 3ddc45508a0b04f3d99199c59769eab710485ed9 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Fran=C3=A7ois=20Pelletier?= <francois@francoispelletier.org>
Date: Sat, 16 Mar 2019 18:40:22 -0400
Subject: [PATCH] predict knn et correction random lecture

---
 Code/Knn.py              | 24 ++++++++++++++++++++----
 Code/entrainer_tester.py |  3 +++
 Code/load_datasets.py    | 23 +++++++++++++++++------
 3 files changed, 40 insertions(+), 10 deletions(-)

diff --git a/Code/Knn.py b/Code/Knn.py
index d63f8f5..a7c8112 100644
--- a/Code/Knn.py
+++ b/Code/Knn.py
@@ -67,9 +67,10 @@ class Knn: #nom de la class à changer
         self.train_labels=train_labels
 
         n,m = train.shape
-        p=m
         nn=np.empty((n,self.k,2))
         
+        self.minkowski_p=m
+        
         # On trouve les k plus proches voisins et leur distance pour chacunes des observations du training set
         # On enlève la valeur testée de la liste des points pour lesquels on mesure la distance car on sait qu'elle vaut 0. 
         # On veut tester sur les autres points seulement
@@ -77,9 +78,9 @@ class Knn: #nom de la class à changer
             i_range = [i for i in range(n)]
             i_range.pop(x)
             nn[x,:,0]=i_range[0:self.k]
-            nn[x,:,1]=np.apply_along_axis(minkowski_distance,1,self.train[i_range[0:self.k]],train[x],p)
+            nn[x,:,1]=np.apply_along_axis(minkowski_distance,1,self.train[i_range[0:self.k]],train[x],self.minkowski_p)
             for i in i_range[self.k:n]:
-                dist = minkowski_distance(self.train[i],train[x],p)
+                dist = minkowski_distance(self.train[i],train[x],self.minkowski_p)
                 nn_dist=nn[x,:,1]
                 distdiff = nn_dist-dist
                 max_distdiff=max(distdiff)
@@ -108,9 +109,24 @@ class Knn: #nom de la class à changer
         alors l'exemple est bien classifié, si non c'est une missclassification
 
         """
+        n,m = self.train.shape
+        nn=np.empty((self.k,2))
+        nn[:,1]=np.apply_along_axis(minkowski_distance,1,self.train[0:self.k],exemple,self.minkowski_p)
+        for i in range(self.k,n):
+            dist = minkowski_distance(self.train[i],exemple,self.minkowski_p)
+            nn_dist=nn[:,1]
+            distdiff = nn_dist-dist
+            max_distdiff=max(distdiff)
+            if(max_distdiff>0):
+                pos_changement = np.argwhere(nn_dist==max(nn_dist))[0]
+                nn[pos_changement,0]=i
+                nn[pos_changement,1]=max_distdiff
+        nn_labels = self.train_labels[nn[:,0].astype(np.int)]
+        nn_mode_label = mode(nn_labels)
         
+        print("Observé:"+str(label)+" Prédit:"+str(nn_mode_label))
         
-        
+        return nn_mode_label
 
     def test(self, test, test_labels):
         """
diff --git a/Code/entrainer_tester.py b/Code/entrainer_tester.py
index 6d7623a..4bc132d 100644
--- a/Code/entrainer_tester.py
+++ b/Code/entrainer_tester.py
@@ -37,6 +37,9 @@ train, train_labels, test, test_labels = ld.load_iris_dataset(0.7)
 
 myKnn.train(train, train_labels)
 
+for i in range(train.shape[0]):
+    myKnn.predict(train[i],train_labels[i])
+
 # Tester votre classifieur
 
 
diff --git a/Code/load_datasets.py b/Code/load_datasets.py
index 29ae4df..b3cd6ea 100644
--- a/Code/load_datasets.py
+++ b/Code/load_datasets.py
@@ -51,13 +51,15 @@ def load_iris_dataset(train_ratio=0.7):
     np_features=np.array(features,dtype=np.float)
     np_labels=np.array(labels,dtype=np.int)
     
-    n_train = int(np_features.shape[0]*train_ratio)
+    n_obs = np_features.shape[0]
     
-    all_indices = [i for i in range(np_features.shape[0])]
+    n_train = int(n_obs*train_ratio)
+    
+    all_indices = [i for i in range(n_obs)]
     random.shuffle(all_indices)
     
     train_index = all_indices[0:n_train]
-    test_index = all_indices[n_train:np_features.shape[0]]
+    test_index = all_indices[n_train:n_obs]
     
     train = np_features[train_index]
     train_labels = np_labels[train_index]
@@ -127,12 +129,21 @@ def load_congressional_dataset(train_ratio):
     np_features=np.array(features,dtype=np.float)
     np_labels=np.array(labels,dtype=np.int)
     
-    train_index = np.random.rand(np_features.shape[0]) < train_ratio
+    n_obs = np_features.shape[0]
+    
+    
+    n_train = int(n_obs*train_ratio)
+    
+    all_indices = [i for i in range(n_obs)]
+    random.shuffle(all_indices)
+    
+    train_index = all_indices[0:n_train]
+    test_index = all_indices[n_train:n_obs]
     
     train = np_features[train_index]
     train_labels = np_labels[train_index]
-    test =  np_features[~train_index]
-    test_labels = np_labels[~train_index] 
+    test =  np_features[test_index]
+    test_labels = np_labels[test_index]
 	
 	# La fonction doit retourner 4 structures de données de type Numpy.
     return (train, train_labels, test, test_labels)