import numpy as np
import matplotlib.pyplot as plt
from pandas import DataFrame
import pandas as pd
import seaborn as sns
# Step 1 - create dataframe
df = pd.read_csv('white_wine_quality.csv',sep=";")
# Step 1 continued. Is the wine acceptable? 1 = True, 0 = False
df['target'] = df['quality'].map( { 0:0,1:0,2:0,3:0,4:0,5:0,6:0,7:1,8:1,9:1,10:1} )
# Step 2. Nan values?
if df.isnull().values.any():
print("Nan values found.")
else:
print("No Nan values found.")
No Nan values found.
# Step 3. Quality groupings and countplot
for i in range(3,11) :
print(f"Quality: {i}, Quality: {len(df[df.quality == i])}")
Quality: 3, Quality: 20
Quality: 4, Quality: 163
Quality: 5, Quality: 1457
Quality: 6, Quality: 2198
Quality: 7, Quality: 880
Quality: 8, Quality: 175
Quality: 9, Quality: 5
Quality: 10, Quality: 0
sns.set_style('whitegrid')
sns.countplot(data=df, x='quality')
# Step 4. For each feature determine if any data instance is an outlier;
# if it is delete that data instance
# Example follows for one feature
print(f"Number of data entires: {len(df)}")
Number of data entires: 4898
# Step 4 example - volatile acidity feature
for feature in df:
f = df['volatile acidity'].values
mean = np.mean(f)
std = np.std(f)
n = len(f)
count = 0
for i in range (0,n):
z = ( f[i] - mean ) / std
if (z>5) :
count = count + 1
df = df.drop( [i])
print ("Number of data instances dropped is ", count)
Number of data instances dropped is 9
Number of data instances dropped is 1
Number of data instances dropped is 1
Number of data instances dropped is 1
Number of data instances dropped is 1
Number of data instances dropped is 1
Number of data instances dropped is 1
Number of data instances dropped is 1
Number of data instances dropped is 1
Number of data instances dropped is 1
Number of data instances dropped is 1
Number of data instances dropped is 1
Number of data instances dropped is 1
# Step 5. get data into correct form
#
df_mod = df.drop(['quality', 'target'], axis=1)
X = df_mod.values
y = df['target'].to_numpy()
print(f"X = {len(X[1])} features x {len(X)} entires")
print(f"Y = {len(y)} data entries")
X = 11 features x 4864 entires
Y = 4864 data entries
# Step 6. Split data in test and trial sets with 80-20 split
from sklearn.model_selection import train_test_split
(X_train, X_test, y_train, y_test) = train_test_split(X, y, test_size=0.2)
print(f"Training: {len(X_train)}, Testing: {len(X_test)}")
Training: 3891, Testing: 973
# Step 7. Scale the data
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
scaler.fit(X_train)
X_train = scaler.transform(X_train)
X_test = scaler.transform(X_test)
# Step 8. Logistic Regression
from sklearn.linear_model import LogisticRegression
lr = LogisticRegression(solver ='lbfgs' )
lr.fit(X_test, y_test)
percent = round(100* lr.score(X_test,y_test), 2)
print(f"Logisitic Regression accuracy = {percent}%")
Logisitic Regression accuracy = 81.09%
# Step 9. Create ANN classifier and train
from sklearn.neural_network import MLPClassifier
mlp = MLPClassifier( max_iter = 300, solver='adam' )
mlp.fit( X_train , y_train )
percent = round(100 * mlp.score( X_test , y_test ), 2)
print (f"ANN accuracy = {percent}%")
ANN accuracy = 83.86%
/shared-libs/python3.7/py/lib/python3.7/site-packages/sklearn/neural_network/_multilayer_perceptron.py:696: ConvergenceWarning: Stochastic Optimizer: Maximum iterations (300) reached and the optimization hasn't converged yet.
ConvergenceWarning,
# Step 10. Create kNN classifier and see what value of k is best
from sklearn.neighbors import KNeighborsClassifier
max = 0
rng = [1, 2, 3, 4, 5]
for i in ( rng ) :
knni = KNeighborsClassifier (n_neighbors = i)
# fitting the data
knni.fit (X_train, y_train)
percent =round(100* knni.score( X_train , y_train ), 2)
print (f"Nearest neighbor {i}, Quality = {percent}%")
if (percent > max ) :
max = percent
index = i
print(f"kNN algorithm with {index} nearest neighbor percent accuracy = {max}%")
Nearest neighbor 1, Quality = 100.0%
Nearest neighbor 2, Quality = 92.78%
Nearest neighbor 3, Quality = 92.06%
Nearest neighbor 4, Quality = 89.31%
Nearest neighbor 5, Quality = 88.49%
kNN algorithm with 1 nearest neighbor percent accuracy = 100.0%
The model with the highest accuracy was the kNN with 1 nearest neighbor. This model obtained 100 percent accuracy while the LR and ANN models were around 81 to 83 percent, making them sub-optimal choices.