[Python] Candidate Elimination Algorithm Implementation Not Working as Expected

I am trying to implement the Candidate Elimination Algorithm to learn a concept from a dataset. However, my code is not producing the expected results. Below is my implementation and the dataset I am using.

My Code:

import pandas as pd
import numpy as np
import os, sys

df = pd.read_csv("chumma.csv")

val_dict = {
"0":["Some","Many"],
"1":['Small','Big','Medium'],
'2':['No'],
'3':['Affordable','Expensive'],
'4':['One','Many','Few']
}

X = df.iloc[:,:-1].values
Y = df.iloc[:,-1].values

S = [['$']*5]
for i in range(len(Y)):
if Y == ' Yes':
S = X.copy()
break

G = [["?"]*len(S)]

def is_consistent(x_data, y_data, g_hyp):
out_hyp = []
for hyp in g_hyp:
is_valid = True
for i in range(len(x_data)):
matches_hyp = True
for j in range(len(hyp)):
if hyp[j]!="?" and hyp[j]!=x_data[j]:
matches_hyp = False
break
if (y_data == ' Yes' and not matches_hyp) or (y_data == ' No' and matches_hyp):
is_valid = False
break

if is_valid and hyp not in out_hyp:
out_hyp.append(hyp)

return out_hyp


for i,instance in enumerate(X):
if Y == ' Yes':
for j in range(len(S)):
if S[j]!=instance[j]:
S[j] = "?"
else:
g_hyp = []
for g in G:
for j in range(len(g)):
if g[j] == "?":
d_vals = val_dict[str(j)]
for k in range(0,len(d_vals)):
if d_vals[k] != instance[j]:
new_g = g.copy()
new_g[j] = d_vals[k]
print(new_g)
g_hyp.append(new_g)
print()
break
G = is_consistent(X[:i+1], Y[:i+1], g_hyp)


print("Final Specific Hypothesis:",S)
print("Final General Hypothesis:",G)


Dataset (chumma.csv):

Citations, Size, InLibrary, Price, Editions, Buy
Some, Small, No, Affordable, One, No
Many, Big, No, Expensive, Many, Yes
Many, Medium, No, Expensive, Few, Yes
Some, Small, No, Affordable, One, No
Many, Big, No, Expensive, Many, Yes
Many, Medium, No, Expensive, Few, Yes


Expected Output The algorithm should output the specific hypothesis (S) and general hypothesis (G) that represent the concept learned from the dataset

Issue The code does not seem to correctly update the specific hypothesis (S) and general hypothesis (G). Specifically:

• S is not being generalized correctly for positive examples.
• G is not being specialized correctly for negative examples.

What is wrong with my implementation of the Candidate Elimination Algorithm and how can I fix the code to correctly update S and G?

Continue reading...