[SML] Misclassification Rates Example Codes: guess, test data

Misclassification Rates Example Codes: guess, test data

포스트 난이도: HOO_Lead

---

# Example Codes

 

# Library
import numpy as np
import pandas as pd
from scipy.stats import norm, t
import matplotlib.pyplot as plt

def getClass1Prop(x,r):

  x=np.array(x)
  dist=np.zeros(len(x_train))

  for i in range(len(x_train)):
    dist[i] = np.linalg.norm(x-x_train[i])
  dist_label_1 = dist[y_train==1]
  dist_1r = dist_label_1[dist_label_1<=r]

  if len(dist[dist<=r]):
    return len(dist_1r)/len(dist[dist<=r])
  else:
    return np.nan
    
def computeMisValData(val_data, r):
  
  x_val = val_data.iloc[:,1:3].to_numpy()
  y_val = val_data['Y'].to_numpy()
  y_pred = np.zeros(len(y_val))

  for i in range(len(x_val)):
    p = getClass1Prop(x_val[i],r)
    y_pred[i] = 1 if p >= 0.5 else 0

  mis_val = (len(y_val)-(y_pred==y_val).sum())/len(y_val)
  return mis_val
  
# check data
df = pd.read_csv("SML.NN.data.csv")
print(df.head(10))
df.info()

# declare train, valid, test
data_train = df[df['set']=='train']
data_valid = df[df['set']== 'valid']
data_test = df[df['set']=='test']

x_train = data_train.iloc[:,1:3].to_numpy()
y_train = data_train['Y'].to_numpy()

x_val = data_valid.iloc[:,1:3].to_numpy()
y_val = data_valid['Y'].to_numpy()

plt.figure()
plt.scatter(x_train[y_train==0][:,0], x_train[y_train==0][:,1],label = "Class 0 of Y")
plt.scatter(x_train[y_train==1][:,0],x_train[y_train==1][:,1],label="Class 1 of Y")
plt.legend()
plt.title('Training Data')
plt.show()

plt.figure()
plt.scatter(x_val[:,0], x_val[:,1])
plt.scatter(x_val[y_val==0][:,0],x_val[y_val==0][:,1],label="class 0 of Y")
plt.scatter(x_val[y_val==1][:,0],x_val[y_val==1][:,1], label="class1 of Y")
plt.legend()
plt.title('Validation Data')
plt.show()

# Compute the misclassification rate (refer to 1.2)
# define compute the misclassification rate
def computeMisRate(dt, r):
  x_test = dt.iloc[:,1:3].to_numpy()
  y_test = dt['Y'].to_numpy()
  y_pred = np.zeros(len(y_test))

  for i in range(len(x_test)):
    p = getClass1Prop(x_test[i],r)
    #proportion higher 0.5
    y_pred[i] = 1 if p >= 0.5 else 0

  misclassificationRate = (len(y_test)-(y_pred==y_test).sum())/len(y_test)
  
  return misclassificationRate
  
  r = np.arange(0.01,1.01,0.01)
mis_total=[]

for i in r:
  mis_total.append(computeMisValData(data_valid, i))

plt.figure()
plt.plot(r,mis_total)
plt.xlabel('r')
plt.ylabel('Misclassification Rate')
plt.show()
rate=r[np.argmin(mis_total)]

print("Guess rate:")
guess = float(input())
print("Lowest misclassification rate is, ", guess ," in my guess by using the plot.")
misclassification = computeMisRate(data_test, rate)
misclassificationGuess = computeMisRate(data_test, guess)
print("r = ",  rate ,  "Misclassification rate(test data): ",  misclassification)
print("r = ",  guess , "Misclassification rate(guess): " , misclassificationGuess)

if (misclassification > misclassificationGuess):
  print("The rate is better than the guess rate.")
else:
  print("Guess rate is better than the misclassification rate.")
  
print("posted by HOO.")

---

# Results

 

Posted by HOO

---

Posted by HOO

---

Posted by HOO

---

Guess rate:
0.23
Lowest misclassification rate is,  0.23  in my guess by using the plot.
r =  0.26 Misclassification rate(test data):  0.045
r =  0.23 Misclassification rate(guess):  0.05
Guess rate is better than the misclassification rate.

---

https://github.com/WhoisHOO/HOOAI/blob/285de860fe9c1675b09998136a752147a510c493/Data%20Science/SML_MisclassificationRate_584.py

GitHub - WhoisHOO/HOOAI: https://whoishoo.tistory.com/

---