# Always suppress warnings to keep everything neat
import warnings
from statsmodels.tools.sm_exceptions import ConvergenceWarning, ValueWarning
warnings.filterwarnings('ignore', category=ValueWarning)
warnings.filterwarnings('ignore', category=ConvergenceWarning)
warnings.filterwarnings('ignore', category=RuntimeWarning)  # e.g., from log/exp on edge values

# Set up key libraries for data, stats, and plots
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
from scipy import stats

sns.set(style='whitegrid')

# Let us simulate flipping a coin 20 times
np.random.seed(42)
flips = np.random.choice(['Heads', 'Tails'], size=20)
print('Coin flips:', flips)

# Let us count the outcomes
heads = np.sum(flips == 'Heads')
tails = np.sum(flips == 'Tails')
print('Heads:', heads)
print('Tails:', tails)

# How unusual is our outcome? Let us compute a p-value.
# We use a binomial test: for 20 flips, expected heads = 10
try:
    p_value = stats.binomtest(heads, n=20, p=0.5, alternative='two-sided')
    # Newer scipy: returns a BinomTestResult object with .pvalue
    print('p-value:', p_value.pvalue)
except AttributeError:
    # Legacy: use stats.binom_test
    pvalue_legacy = stats.binom_test(heads, n=20, p=0.5, alternative='two-sided')
    print('p-value:', pvalue_legacy)
    # for lesson, make sure p_value is defined & compatible
    class Dummy:
        def __init__(self, p): self.pvalue = p
    p_value = Dummy(pvalue_legacy)
    

# What does the p-value mean?
if p_value.pvalue < 0.05:
    print('We reject the null hypothesis. The coin may not be fair!')
else:
    print('We fail to reject the null. The coin seems fair enough.')
    

# See errors in action: let us simulate many experiments
n_exp = 1000
false_alarms = 0
for i in range(n_exp):
    flips_sim = np.random.choice(['Heads', 'Tails'], size=20)
    heads_sim = np.sum(flips_sim == 'Heads')
    try:
        p_sim = stats.binomtest(heads_sim, n=20, p=0.5, alternative='two-sided')
        pval = p_sim.pvalue
    except AttributeError:
        pval = stats.binom_test(heads_sim, n=20, p=0.5, alternative='two-sided')
    if pval < 0.05:
        false_alarms += 1
print('False alarms (Type I errors) in 1000 simulated fair coins:', false_alarms)

# Data setup: Load the tips dataset
tips = sns.load_dataset('tips')
print('Shape:', tips.shape)
tips.head()

# Compare tips on Sunday vs. other days: plot and summary
sunday = tips[tips['day'] == 'Sun']['tip']
other = tips[tips['day'] != 'Sun']['tip']
plt.figure(figsize=(7,4))
sns.histplot(sunday, color='red', kde=True, label='Sunday', bins=12)
sns.histplot(other, color='blue', kde=True, label='Other Days', bins=12)
plt.legend()
plt.title('Tip distributions: Sunday vs. Other Days')
plt.xlabel('Tip Amount ($)')
plt.ylabel('Frequency')
plt.show()

# Hypothesis test: Are Sunday tips different?
t_stat, p_tip = stats.ttest_ind(sunday, other, equal_var=False)
print('t-statistic:', t_stat)
print('p-value:', p_tip)

# Interpreting our result
if p_tip < 0.05:
    print('Tips on Sundays are likely different from other days!')
else:
    print('No significant difference found in average tips.')
    

# Short practice: Enter tip amounts for two groups and test
group1 = []
group2 = []
print('Enter 5 tip amounts for Group 1:')
for i in range(5):
    group1.append(float(input(f'Tip {i+1} for Group 1: ')))
print('Enter 5 tip amounts for Group 2:')
for i in range(5):
    group2.append(float(input(f'Tip {i+1} for Group 2: ')))
t_stat2, p_value2 = stats.ttest_ind(group1, group2, equal_var=False)
print('t-statistic:', t_stat2)
print('p-value:', p_value2)

# Practice: Interpret your result
if p_value2 < 0.05:
    print('Your groups have a real difference in tips!')
else:
    print('No major difference in tips between your groups.')