I'm following sentdex's youtube channel ML tutorial.

So as I was coding along on how to build your own KNN algorithm, I noticed that my accuracy was very low, in the 60s almost every time. I had made a few changes, but then I used his code line by line, and the same dataset, yet somehow he gets accuracies in the range of 95-98%, while mine is 60-70%. I'm really not able to figure out the reason behind such a huge difference.

I also have a second question which has to do with the confidence of the predictions. The value of the confidence is supposed to be within 0-1 right? But for me, they're all identical, and in the 70s. Let me explain with a screenshot

My code:

# Importing libraries
import numpy as np
import pandas as pd
from collections import Counter
import warnings
import random

# Algorithm

def k_nearest(data,predict,k=5):
    if len(data)>=k:
        warnings.warn("stupid, your data has more dimensions than prescribed")
    distances = []
    for group in data: # The groups of 2s and 4s
        for features in data[group]: # values in 2 and 4 respectively
            #euclidean_distance = np.sqrt(np.sum((np.array(features) - np.sum(predict)) **2))
            euclidean_distance = np.linalg.norm(np.array(features) - np.array(predict))
            distances.append([euclidean_distance,group])
    votes = [i[1] for i in sorted(distances)] # adding the sorted(ascending) group names
    votes_result = Counter(votes).most_common(1)[0][0] # the most common element
    confidence = float((Counter(votes).most_common(1)[0][1]))/float(k)#ocuurences of the most common element
    
    return votes_result,confidence
#reading the data
df = pd.read_csv("breast_cancer.txt")
df.replace("?",-99999,inplace=True)
#df.replace("?", np.nan,inplace=True)
#df.dropna(inplace=True)
df.drop("id",axis = 1,inplace=True)
full_data = df.astype(float).values.tolist() # Converting to list because our function is written like that
random.shuffle(full_data)
#print(full_data[:10])

test_size = 0.2
train_set = {2:[],4:[]}
test_set = {2:[],4:[]}
train_data = full_data[:-int(test_size*len(full_data))] # Upto the last 20% of the og dateset
test_data = full_data[-int(test_size*len(full_data)):] # The last 20% of the dataset

# Populating the dictionary
for i in train_data:
    train_set[i[-1]].append(i[:-1]) # appending with features and leaving out the label
for i in test_data:
    test_set[i[-1]].append(i[:-1]) # appending with features and leaving out the label

# Testing 
correct,total = 0,0
for group in test_set:
    for data in test_set[group]:
        vote,confidence = k_nearest(train_set, data,k=5)
        if vote == group:
            correct  =1
        else:
            print(confidence)
        total  = 1 
print("Accuracy is",correct/total)

Link to the dataset breast-cancer-wisconsin.data

CodePudding user response：

There's a mistake in your k_nearest function, you need to return only the top k of distances, not the whole list. So it should be:

votes = [i[1] for i in sorted(distances)[:k]]

Instead of in your code:

votes = [i[1] for i in sorted(distances)]

We can rewrite your function:

def k_nearest(data,predict,k=5):
    distances = []
    for group in data: 
        for features in data[group]: 
            euclidean_distance = np.linalg.norm(np.array(features) - np.array(predict))
            distances.append([euclidean_distance,group])
    votes = [i[1] for i in sorted(distances)[:k]] 
    votes_result = Counter(votes).most_common(1)[0][0] 
    confidence = float((Counter(votes).most_common(1)[0][1]))/float(k)

return votes_result,confidence

And run your code, I am not so sure about replacing "?" with -999 so I read it in as na :

import pandas as pd
from collections import Counter
import random
import numpy as np

url = 'https://archive.ics.uci.edu/ml/machine-learning-databases/breast-cancer-wisconsin/breast-cancer-wisconsin.data'
df = pd.read_csv(url,header=None,na_values="?")
df = df.dropna()
full_data = df.iloc[:,1:].astype(float).values.tolist() 
random.seed(999)
random.shuffle(full_data)

test_size = 0.2
train_set = {2:[],4:[]}
test_set = {2:[],4:[]}
train_data = full_data[:-int(test_size*len(full_data))] 
test_data = full_data[-int(test_size*len(full_data)):] 

for i in train_data:
    train_set[i[-1]].append(i[:-1]) 
for i in test_data:
    test_set[i[-1]].append(i[:-1]) 

correct,total = 0,0
for group in test_set:
    for data in test_set[group]:
        vote,confidence = k_nearest(train_set, data,k=5)
        if vote == group:
            correct  =1
        else:
            print(confidence)
        total  = 1 
print("Accuracy is",correct/total)

Gives:

1.0
0.8
1.0
0.6
0.6
0.6
0.6
Accuracy is 0.9485294117647058