import warnings
warnings.filterwarnings("ignore")  # Silence any warnings for a clean start

# First, let us import the libraries we will use
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

print("Ready to go!")

# Data setup: Load a real sales dataset
url = "https://raw.githubusercontent.com/jbrownlee/Datasets/master/shampoo.csv"
data = pd.read_csv(url)
print("Rows, columns:", data.shape)
data.head()

# Let us see all column names
print(data.columns.tolist())

# Plot the sales data to understand its pattern
plt.figure(figsize=(8, 4))
plt.plot(data['Sales'], marker='o')
plt.title('Monthly Shampoo Sales')
plt.xlabel('Month')
plt.ylabel('Sales')
plt.grid(True)
plt.show()

# Simple exponential smoothing: apply to sales data
from statsmodels.tsa.holtwinters import SimpleExpSmoothing

sales = data['Sales'].astype(float)
model = SimpleExpSmoothing(sales)
fit = model.fit(smoothing_level=0.4, optimized=False)
pred = fit.fittedvalues

plt.figure(figsize=(8,4))
plt.plot(sales, label='Original')
plt.plot(pred, label='Smoothed', color='red')
plt.legend()
plt.title('Simple Exponential Smoothing')
plt.show()

# Try a different smoothing level
fit2 = model.fit(smoothing_level=0.9, optimized=False)
pred2 = fit2.fittedvalues

plt.figure(figsize=(8,4))
plt.plot(sales, label='Original')
plt.plot(pred2, color='green', label='Level 0.9')
plt.plot(pred, color='red', label='Level 0.4')
plt.legend()
plt.title('Compare Smoothing Levels')
plt.show()

# Holt: Handles trends in data
from statsmodels.tsa.holtwinters import ExponentialSmoothing

holt_model = ExponentialSmoothing(sales, trend='add', seasonal=None)
holt_fit = holt_model.fit(optimized=True)
holt_pred = holt_fit.fittedvalues

plt.figure(figsize=(8,4))
plt.plot(sales, label='Original')
plt.plot(holt_pred, color='purple', label='Holt Trend')
plt.legend()
plt.title('Holt Method for Trend')
plt.show()

# Holt-Winters: handles both trend and seasonality
hw_model = ExponentialSmoothing(sales, trend='add', seasonal='add', seasonal_periods=12)
hw_fit = hw_model.fit(optimized=True)
hw_pred = hw_fit.fittedvalues

plt.figure(figsize=(8,4))
plt.plot(sales, label='Original')
plt.plot(hw_pred, color='orange', label='Holt-Winters')
plt.legend()
plt.title('Holt-Winters (Trend + Seasonality)')
plt.show()

# Make a forecast for future months
future = hw_fit.forecast(6)
print('Next 6 months sales forecast:')
print(future)

# Plot prediction vs. real data for the last points
plt.figure(figsize=(8,4))
plt.plot(sales, label='Actual Sales')
plt.plot(hw_fit.fittedvalues, label='Fitted', color='orange')
plt.plot(range(len(sales), len(sales)+6), future, 'go--', label='Forecast')
plt.legend()
plt.title('Real vs. Fitted vs. Forecasted Sales')
plt.show()

# Safe access: What if there is missing data?
print(data.isnull().sum())

# Fill missing values with the previous month
data_filled = data.fillna(method='ffill')
print(data_filled.isnull().sum())

# Quick mini-project: Predict sales after user input
user_months = int(input('How many months to predict? (1-12): '))
my_forecast = hw_fit.forecast(user_months)
print('Here is your forecast:')
print(my_forecast)

# Troubleshooting: What if fitting fails?
try:
    broken_model = ExponentialSmoothing([], trend='add', seasonal='add', seasonal_periods=12)
    broken_fit = broken_model.fit()
except Exception as e:
    print('Error:', e)
    

# Extra tip: You can save your fitted model
import pickle
with open('hw_model.pkl', 'wb') as f:
    pickle.dump(hw_fit, f)
print('Model saved!')

# Challenge: Try your own smoothing level
level = float(input('Pick a smoothing level between 0 and 1: '))
user_fit = model.fit(smoothing_level=level, optimized=False)
print('Your smoothed values:')
print(user_fit.fittedvalues.head())