LinearRegressionAndRegularization.py

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from warnings import filterwarnings
filterwarnings('ignore')
pd.set_option('display.max_columns', None)
pd.set_option('display.max_colwidth', None)
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LinearRegression,Lasso,LassoCV,Ridge,RidgeCV,ElasticNet,ElasticNetCV
from sklearn.metrics import mean_absolute_error,r2_score
import pickle

# Create Dataframe and Read the dataset using Pandas
df = pd.read_csv('./data/Algerian_forest_fires_dataset_CLEANED.csv')
print(df.head())
print(df.columns)
df.drop(['day', 'month', 'year'],axis=1,inplace=True)
# print(df.head())
print(df['Classes'].value_counts())
df['Classes']=np.where(df['Classes'].str.contains("not fire"),0,1)
print(df['Classes'].value_counts())
print(df.tail())
# Independent and dependent features, Let's calculate Fire Weather Index (0 to 31.1)
X=df.drop('FWI',axis=1)
y=df["FWI"]
print(X.head())
print(y.head())

# Train test split
X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.25,random_state=42)
print(X_train.shape,X_test.shape)
# Feature selection based on correlation
print(X_train.corr())

# If independent features are highly correlated, we need to eliminate those features, both features will be conveying same info==>Multicollinearity
# check for multicollinearity
plt.figure(figsize=(12,10))
corr=X_train.corr()
sns.heatmap(corr,annot=True)
plt.show()
def correlation(dataset,threshold):
    col_corr=set()
    corr_matrix=dataset.corr()
    for i in range(len(corr_matrix.columns)):
        for j in range(i):
            if abs(corr_matrix.iloc[i,j])>threshold:
                colname=corr_matrix.columns[i]
                col_corr.add(colname)
    return col_corr
# threshold-domain expert
corr_feature=correlation(X_train,0.85)
print(corr_feature)
# Drop features when correlation is > threshold
X_train.drop(corr_feature,axis=1,inplace=True)
X_test.drop(corr_feature,axis=1,inplace=True)
print(X_train.shape,X_test.shape)

# Feature scaling or Standardization
scaler=StandardScaler()
Xtrains=scaler.fit_transform(X_train)
Xtests=scaler.transform(X_test)
print(Xtrains)

# Box plot to understand effect of standardization
plt.subplots(figsize=(15,5))
plt.subplot(121)
sns.boxplot(data=X_train)
plt.title('X_train before scaling')
plt.subplot(122)
sns.boxplot(data=Xtrains)
plt.title('X_train after scaling')
plt.show()

# Linear Regression model
linreg=LinearRegression()
linreg.fit(Xtrains,y_train)
ypred=linreg.predict(Xtests)
mae=mean_absolute_error(y_test,ypred)
score=r2_score(y_test,ypred)
print('mae:',mae,', r2 score:',score)
plt.scatter(y_test,ypred)
# plt.show()

# Lasso Regression
lasso=Lasso()
lasso.fit(Xtrains,y_train)
ypred=lasso.predict(Xtests)
mae=mean_absolute_error(y_test,ypred)
score=r2_score(y_test,ypred)
print('mae:',mae,', r2 score:',score)
plt.scatter(y_test,ypred,c='red')
# plt.show()

# Cross Validation Lasso
lassocv=LassoCV(cv=5,random_state=0)
lassocv.fit(Xtrains,y_train)
ypred=lassocv.predict(Xtests)
mae=mean_absolute_error(y_test,ypred)
score=r2_score(y_test,ypred)
print('mae:',mae,', r2 score:',score)
plt.scatter(y_test,ypred,c='green')
# plt.show()
print('*****************************************')
# print(lassocv.alpha_,lassocv.alphas_,lassocv.mse_path_)

# Ridge Regression
ridge=Ridge()
ridge.fit(Xtrains,y_train)
ypred=ridge.predict(Xtests)
mae=mean_absolute_error(y_test,ypred)
score=r2_score(y_test,ypred)
print('mae:',mae,', r2 score:',score)
plt.scatter(y_test,ypred,c='yellow')
# plt.show()

# Ridge Cross validation
ridgecv=RidgeCV(cv=7)
ridgecv.fit(Xtrains,y_train)
ypred=ridgecv.predict(Xtests)
mae=mean_absolute_error(y_test,ypred)
score=r2_score(y_test,ypred)
print('mae:',mae,', r2 score:',score)
plt.scatter(y_test,ypred,c='pink')
plt.show()
print(ridgecv.get_params())

# ElasticNEt Regression
elast=ElasticNet()
elast.fit(Xtrains,y_train)
ypred=elast.predict(Xtests)
mae=mean_absolute_error(y_test,ypred)
score=r2_score(y_test,ypred)
print('mae:',mae,', r2 score:',score)
plt.scatter(y_test,ypred,c='yellow')
# plt.show()

# elasticnet Cross validation
elastcv=ElasticNetCV(cv=7)
elastcv.fit(Xtrains,y_train)
ypred=elastcv.predict(Xtests)
mae=mean_absolute_error(y_test,ypred)
score=r2_score(y_test,ypred)
print('mae:',mae,', r2 score:',score)
plt.scatter(y_test,ypred,c='pink')
plt.show()
# print(elastcv.get_params())

# Pickle the machine learning models, preprocessing model standardscaler
pickle.dump(scaler,open(r'./data/scaler.pkl','wb'))
pickle.dump(ridge,open(r'./data/ridge.pkl','wb'))