Linear_Regression_code.py

# -*- coding: utf-8 -*-
"""Untitled13.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1mVogyEwZ56lzTCW9NevzCxC-ewNG_jSP
"""

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import warnings
warnings.filterwarnings('ignore')
from sklearn.linear_model import LinearRegression
from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import f_regression
from sklearn.metrics import mean_absolute_error as mae
from sklearn.metrics import mean_squared_error as rmse
from sklearn.metrics import mean_absolute_percentage_error as mape

df = pd.read_csv('/content/data_sales.csv')
df.head(10)

df.info()

df.drop(['f0', 'f1'], axis=1, inplace=True)

df.sample(5)

df['date'] = pd.to_datetime(df['date'], format='%d-%m-%Y')

df.describe()

df = df.set_index('date').interpolate()

df = df.reset_index(drop=False)
df.sample(5)

df.sort_values(by='date', ascending = True, inplace = True)

df = df[df['store'] == 1]
df = df[df['item'] == 1]

df['date'] = pd.to_datetime(df['date'], format='%Y-%m-%d') # convert date column to datatime object

# Create Date-related Features to be used for EDA and Supervised ML: Regression
df['year'] = df['date'].dt.year
df['month'] = df['date'].dt.month
df['day'] = df['date'].dt.day
df['weekday'] = df['date'].dt.weekday
df['weekday'] = np.where(df.weekday == 0, 7, df.weekday)

# Split the series to predict the last 3 months of 2017
temp_df = df.set_index('date')
train_df = temp_df.loc[:'2017-09-30'].reset_index(drop=False)
test_df = temp_df.loc['2017-10-01':].reset_index(drop=False)

train_df.head()

for i in range(1,8):
    lag_i = 'lag_' + str(i)
    df[lag_i] = df.sales.shift(i)

# Rolling window
df['rolling_mean'] = df.sales.rolling(window=7).mean()
df['rolling_max'] = df.sales.rolling(window=7).max()
df['rolling_min'] = df.sales.rolling(window=7).min()

df = df.dropna(how='any', inplace=False)
df = df.drop(['store', 'item'], axis=1)

# Split the series to predict the last 3 months of 2017
df = df.set_index('date')
reg_train_df = df.loc[:'2017-09-30']
reg_test_df= df.loc['2017-10-01':]

X_train = reg_train_df.drop(['sales'], axis=1)
y_train = reg_train_df['sales'].values

X_test = reg_test_df.drop(['sales'], axis=1)
y_test = reg_test_df['sales'].values

#Univariate SelectKBest class to extract top 5 best features
top_features = SelectKBest(score_func=f_regression, k=5)
fit = top_features.fit(X_train, y_train)
df_scores = pd.DataFrame(fit.scores_)
df_columns = pd.DataFrame(X_train.columns)

#concat two dataframes for better visualization
feature_scores = pd.concat([df_columns, df_scores], axis=1)
feature_scores.columns = ['Feature','Score']  # naming the dataframe columns
print(feature_scores.nlargest(5,'Score'))

corr = reg_train_df.corr()
fig = plt.figure(figsize=(10,7))
sns.heatmap(corr, linewidths=.5)

fig, axs = plt.subplots(ncols=2, figsize=(14,7))
sns.scatterplot(x=reg_train_df['rolling_mean'], y=reg_train_df['sales'], ax=axs[0])
axs[0].set(title='Linear relationship between sales and rolling_mean of sales')
sns.scatterplot(x=reg_train_df['rolling_max'], y=reg_train_df['sales'], ax=axs[1])
axs[1].set(title='Linear relationship between sales and rolling_max of sales')

fig, axs = plt.subplots(ncols=2, figsize=(14,7))
sns.scatterplot(x=reg_train_df['rolling_min'], y=reg_train_df['sales'], ax=axs[0])
axs[0].set(title='Linear relationship between sales and rolling_min of sales')
sns.scatterplot(x=reg_train_df['lag_7'], y=reg_train_df['sales'], ax=axs[1])
axs[1].set(title='Linear relationship between sales and lag_7 of sales')

plt.show()

X_train = X_train[['rolling_mean', 'rolling_max', 'rolling_min', 'lag_7', 'lag_1']]
X_test = X_test[['rolling_mean', 'rolling_max', 'rolling_min', 'lag_7', 'lag_1']]

# fit model
model = LinearRegression()
model.fit(X_train, y_train)

preds = model.predict(X_test)

errors_df = reg_test_df[['sales']]
errors_df['pred_sales'] = preds
errors_df['errors'] = preds - y_test
errors_df.insert(0, 'model', 'LinearRegression')

# eval predictions
fig = plt.figure(figsize=(14,7))
plt.plot(reg_train_df.index, reg_train_df['sales'], label='Train')
plt.plot(reg_test_df.index, reg_test_df['sales'], label='Test')
plt.plot(errors_df.index, errors_df['pred_sales'], label='Forecast - Linear Regression')
plt.legend(loc='best')
plt.xlabel('date')
plt.ylabel('sales')
plt.title('Forecasts using Linear Regression model')
plt.show()

fig = plt.figure(figsize=(14,7))
plt.plot(errors_df.index, errors_df.errors, label='errors')
plt.plot(errors_df.index, errors_df.sales, label='actual sales')
plt.plot(errors_df.index, errors_df.pred_sales, label='forecast')
plt.legend(loc='best')
plt.xlabel('date')
plt.ylabel('sales')
plt.title('Linear Regression forecasts with actual sales and errors')
plt.show()