first commit

KNN.py

@@ -0,0 +1,68 @@
+import numpy as np
+import load_csv
+
+def distance_L1(x1,x2):
+  # x1 and x2 can be in different shapes on the sample dim
+  # if one of them is the test point and the other is the whole training set, then the minimum distance is in the returned distance array
+  x1=np.array(x1)
+  x2=np.array(x2)
+  return np.sum(np.abs(x1-x2),axis=-1)
+
+def distance_L2(x1,x2):
+  x1=np.array(x1)
+  x2=np.array(x2)
+  return np.sqrt(np.sum(np.power(x1-x2,2),axis=-1))
+
+def std_normalization(inputs,mean_std=None,axis=0):
+  if mean_std==None:
+    mean=np.mean(inputs,axis=axis)
+    std=np.std(inputs,axis=axis)
+    return [mean,std,(inputs-mean)/std]
+  else:
+    mean=mean_std[0]
+    std=mean_std[1]
+    return [mean,std,(inputs-mean)/std]
+
+def scaling(inputs,axis=0):
+  # scale inputs to 0-1 
+  return inputs/np.amax(inputs,axis=axis)
+
+def KNN(x,train,K=3):
+  # train should have one more column than x as the label column
+  train=np.array(train)
+  x=np.array(x)
+  train_inputs=train[:,:-1]
+
+  [mean,std,train_inputs]=std_normalization(train_inputs)
+  [_,_,x]=std_normalization(x,mean_std=[mean,std])
+  dist=distance_L2(x,train_inputs)
+  sorted_indices=np.argsort(dist,axis=0)
+  print(sorted_indices[0])
+  bin_results=np.bincount(np.int32(train[sorted_indices[:K]][-1]))
+  return np.float32(np.argmax(bin_results))
+
+def acc(pred):
+  answer=load_csv.csv_reader("./pima/test_answer.csv")
+  correct=0
+  for pred_row,answer_row in zip(pred[1:],answer[1:]):
+    if answer_row[1]==pred_row[1]:
+      correct+=1
+  return correct/(len(answer)-1)
+
+
+if __name__=="__main__":
+  train_load=load_csv.csv_reader("./pima/train.csv")
+  val_load=load_csv.csv_reader("./pima/val.csv")
+  test_load=load_csv.csv_reader("./pima/test.csv")
+
+  train=np.array(train_load[1:])
+  train=train[:,1:]
+
+  pred=[["ID","Outcome"]]  
+  for i in range(len(test_load[1:])):
+    pred.append([test_load[i+1][0],KNN(test_load[i+1][1:],train,K=19)])
+  print(acc(pred))
+  load_csv.csv_writer(pred,"./pima/KNN_pred.csv")
+
+
+