cross_validation.py

#file name: cross_validation.py
"""Set up the dataset"""
import pandas as pd
df = pd.read_csv('Tunnel_1_4Linear8Sensors9ClassesCappedRange8Fast_10.txt', sep='\t')
from sklearn.model_selection import train_test_split
train_X, val_X, train_y, val_y = train_test_split(df.loc[:,df.columns != "Class"], df.Class, random_state = 0)

"""Use cross-validation on the training set to find out the best value for k (in an arbitrary range of k)."""
from sklearn.model_selection import cross_val_score
from sklearn.utils import shuffle
from sklearn.neighbors import KNeighborsClassifier
import pandas as pd
import numpy as np
shuffled = shuffle(pd.concat([train_X, train_y], axis=1))
for k in range(1,7):
  classifier = KNeighborsClassifier(n_neighbors=k)
  score = cross_val_score(classifier, shuffled.loc[:,shuffled.columns != "Class"], shuffled.Class, cv=10)
  #print(score)
  print(np.median(score))
"""
0.9505526451965509
0.9402730375426621
0.9386145205272369
0.93259385665529
0.924123936717619
0.9148211243611585
0.9028960817717206
"""
#We see that the highest accuracy is reached for k=1 (in the first line).
#Now we test how good 1NN performs on the testing data set.
knn = KNeighborsClassifier(n_neighbors=1)
knn.fit(train_X, train_y)
predictions = knn.predict(val_X)
accuracy = sum(predictions == val_y)/len(predictions)
print(accuracy)
"""
0.9524296675191816
"""