The Full Statistical Picture
python
import pandas as pd, numpy as np
np.random.seed(42)
df = pd.read_csv("sales.csv") # or use sample data
sales = pd.Series(np.random.normal(52000,15000,200).round(0))
# Complete summary
print(sales.describe())
print(f"\nSkewness: {sales.skew():.3f}")
print(f"Kurtosis: {sales.kurtosis():.3f}")
print(f"CV: {sales.std()/sales.mean()*100:.1f}%")
Output
count 200.000
mean 52,156.000
std 15,284.000
min 9,781.000
25% 41,855.000
50% 52,244.000
75% 62,726.000
max 93,912.000
Skewness: 0.048
Kurtosis: -0.098
CV: 29.3%Correlation Matrix
python
import pandas as pd, numpy as np
np.random.seed(42)
n = 100
df = pd.DataFrame({
"Sales": np.random.normal(50000,10000,n),
"Marketing": np.random.normal(5000,1000,n),
"Staff": np.random.randint(5,50,n),
"Season": np.random.choice([1,2,3,4],n),
})
# Add correlation
df["Sales"] += df["Marketing"] * 4
corr = df.corr()
print(corr.round(2))
# Find top correlated pairs
import numpy as np
mask = np.triu(np.ones(corr.shape), k=1).astype(bool)
pairs = corr.where(mask).stack().sort_values(ascending=False)
print("\nTop correlations:")
print(pairs.head(3))
🎯 TASKPractice: Descriptive Statistics
Perform a complete descriptive analysis on a sales dataset: central tendency, spread, shape, and identify key insights.
Starter code:
python
import pandas as pd, numpy as np
np.random.seed(42)
df = pd.DataFrame({
"Region": np.random.choice(["Cairo","Alex","Giza","Aswan"],200),
"Product": np.random.choice(["Laptop","Phone","Tablet"],200),
"Sales": np.random.lognormal(10.8, 0.4, 200).round(0),
"Units": np.random.randint(1,15,200),
})
print("=== Overall Sales Statistics ===")
print(f"Mean: {df['Sales'].mean():>12,.0f}")
print(f"Median: {df['Sales'].median():>12,.0f}")
print(f"Std Dev: {df['Sales'].std():>12,.0f}")
print(f"Skew: {df['Sales'].skew():>12.3f}")
print("\n=== By Region ===")
by_region = df.groupby("Region")["Sales"].agg(
Mean="mean", Median="median", Std="std", Count="count"
).round(0)
print(by_region.sort_values("Mean", ascending=False))
print("\n=== Correlation ===")
print(df[["Sales","Units"]].corr().round(3))
Basic GroupBy
python
import pandas as pd, numpy as np
np.random.seed(42)
df = pd.DataFrame({
"Region": np.random.choice(["Cairo","Alex","Giza"],100),
"Product": np.random.choice(["Laptop","Phone","Tablet"],100),
"Sales": np.random.randint(5000,80000,100),
"Units": np.random.randint(1,10,100),
})
# Group by one column
print(df.groupby("Region")["Sales"].sum().sort_values(ascending=False))
# Multiple aggregations
summary = df.groupby("Region")["Sales"].agg(
Total="sum", Average="mean", Count="count", Max="max"
).round(0)
print(summary)
Output
Region
Cairo 1,245,000
Alex 987,000
Giza 823,000
Name: Sales, dtype: int64
Total Average Count Max
Region
Alex 987000 28229.0 35 79000
Cairo 1245000 35571.0 35 79000
Giza 823000 27433.0 30 77000Named Aggregations & Transform
python
# Group by multiple columns
pivot = df.groupby(["Region","Product"])["Sales"].agg(["sum","mean"]).round(0)
print(pivot.head(6))
# transform — add group stats to each row
df["Region_Avg"] = df.groupby("Region")["Sales"].transform("mean")
df["vs_avg"] = df["Sales"] - df["Region_Avg"]
df["Above_Avg"] = df["vs_avg"] > 0
# Filter groups (only regions with avg > 30000)
high_regions = df.groupby("Region").filter(lambda x: x["Sales"].mean() > 30000)
print(f"High regions: {high_regions['Region'].unique()}")
🎯 TASKPractice: GroupBy — Grouped Analysis
Analyze sales by region AND product. Find top product per region, regions missing from top 10 by revenue, and flag underperforming combinations.
Starter code:
python
import pandas as pd, numpy as np
np.random.seed(42)
df = pd.DataFrame({
"Region": np.random.choice(["Cairo","Alex","Giza","Aswan","Luxor"],200),
"Product": np.random.choice(["Laptop","Phone","Tablet","Monitor"],200),
"Sales": np.random.randint(3000,80000,200),
"Quarter": np.random.choice(["Q1","Q2","Q3","Q4"],200),
})
# Top product per region
by_rp = df.groupby(["Region","Product"])["Sales"].sum()
top_product = by_rp.groupby("Region").idxmax().apply(lambda x: x[1])
print("Top product per region:")
print(top_product)
# Overall totals per region
region_totals = df.groupby("Region")["Sales"].sum().sort_values(ascending=False)
print("\nRegion totals:")
print(region_totals)
# Underperforming: below 25th percentile for their region
df["Region_Q25"] = df.groupby("Region")["Sales"].transform(lambda x: x.quantile(0.25))
weak = df[df["Sales"] < df["Region_Q25"]][["Region","Product","Sales"]].head(5)
print("\nUnderperforming sales (sample):")
print(weak.to_string(index=False))
Types of Joins
python
import pandas as pd
orders = pd.DataFrame({
"Order_ID": [1,2,3,4,5],
"Customer_ID":[101,102,101,103,999],
"Sales": [5000,8000,3000,12000,7000],
})
customers = pd.DataFrame({
"Customer_ID":[101,102,103,104],
"Name": ["Ahmed","Sara","Omar","Layla"],
"City": ["Cairo","Alex","Giza","Cairo"],
})
# Left join — all orders, match customers
left = pd.merge(orders, customers, on="Customer_ID", how="left")
# Inner join — only matches
inner = pd.merge(orders, customers, on="Customer_ID", how="inner")
# Outer join — all from both
outer = pd.merge(orders, customers, on="Customer_ID", how="outer")
print(f"Orders: {len(orders)}")
print(f"Left: {len(left)}")
print(f"Inner: {len(inner)}")
print(f"Outer: {len(outer)}")
print(left)
Output
Orders: 5
Left: 5
Inner: 4
Outer: 6
Order_ID Customer_ID Sales Name City
0 1 101 5000 Ahmed Cairo
1 2 102 8000 Sara Alex
2 3 101 3000 Ahmed Cairo
3 4 103 12000 Omar Giza
4 5 999 7000 NaN NaNMerging on Different Column Names
python
# Different column names in each DataFrame
orders2 = pd.DataFrame({"OrdID":[1,2,3],"CustID":[101,102,103],"Sales":[5000,8000,3000]})
customers2 = pd.DataFrame({"ID":[101,102,103],"Name":["Ahmed","Sara","Omar"]})
merged = pd.merge(orders2, customers2,
left_on = "CustID",
right_on = "ID",
how = "left"
)
print(merged)
💡 Tip: Always check the row count after merging — an unexpected increase means duplicate keys in one of the DataFrames.
🎯 TASKPractice: Merging DataFrames
Merge 3 tables: orders, customers, and products. Calculate revenue per customer and identify top customers by city.
Starter code:
python
import pandas as pd
orders = pd.DataFrame({
"Order_ID": range(1,11),
"Cust_ID": [1,2,1,3,2,4,3,1,5,2],
"Prod_ID": [101,102,103,101,102,103,101,102,103,101],
"Quantity": [2,1,3,1,2,1,2,1,1,3],
})
customers = pd.DataFrame({
"Cust_ID": [1,2,3,4,5],
"Name": ["Ahmed","Sara","Omar","Layla","Karim"],
"City": ["Cairo","Alex","Cairo","Giza","Alex"],
})
products = pd.DataFrame({
"Prod_ID": [101,102,103],
"Name": ["Laptop","Phone","Tablet"],
"Price": [12000,6500,4200],
})
# Merge all three
merged = (orders
.merge(customers, on="Cust_ID")
.merge(products, on="Prod_ID", suffixes=("","_prod"))
)
merged["Revenue"] = merged["Quantity"] * merged["Price"]
# Top customers
top_cust = merged.groupby(["Name","City"])["Revenue"].sum().sort_values(ascending=False)
print("Revenue by customer:")
print(top_cust)
print("\nRevenue by city:")
print(merged.groupby("City")["Revenue"].sum().sort_values(ascending=False))
pd.concat — Stack DataFrames
python
import pandas as pd
# Vertical stack (add rows)
q1 = pd.DataFrame({"Month":["Jan","Feb","Mar"],"Sales":[45000,52000,48000]})
q2 = pd.DataFrame({"Month":["Apr","May","Jun"],"Sales":[61000,58000,72000]})
all_data = pd.concat([q1, q2], ignore_index=True)
print(all_data)
# Horizontal stack (add columns)
names = pd.DataFrame({"Name":["Ahmed","Sara","Omar"]})
scores = pd.DataFrame({"Score":[85,92,78],"Grade":["B","A","C"]})
combined = pd.concat([names, scores], axis=1)
print(combined)
Output
Month Sales
0 Jan 45000
1 Feb 52000
2 Mar 48000
3 Apr 61000
4 May 58000
5 Jun 72000
Name Score Grade
0 Ahmed 85 B
1 Sara 92 A
2 Omar 78 CCombining Multiple Files
python
import pandas as pd, glob, os
# Load and combine multiple CSV files
# files = glob.glob("sales_*.csv")
# dfs = [pd.read_csv(f) for f in files]
# all = pd.concat(dfs, ignore_index=True)
# Add source column to track origin
dfs = [q1.assign(quarter="Q1"), q2.assign(quarter="Q2")]
all_q = pd.concat(dfs, ignore_index=True)
print(all_q)
# Stack different-shaped DataFrames (union of columns)
df_a = pd.DataFrame({"A":[1,2],"B":[3,4]})
df_b = pd.DataFrame({"B":[5,6],"C":[7,8]})
stacked = pd.concat([df_a, df_b], ignore_index=True)
print(stacked) # NaN where columns don't exist
💡 Tip: Add a source column before concat to track which file/period each row came from. Essential when combining multiple sheets or files.
🎯 TASKPractice: Concat & Join
Combine 4 quarterly DataFrames into one annual dataset. Add quarter labels, verify totals, and create a quarterly comparison.
Starter code:
python
import pandas as pd, numpy as np
np.random.seed(42)
quarters = []
for q in range(1,5):
df = pd.DataFrame({
"Month": [f"{(q-1)*3+m:02d}" for m in range(1,4)],
"Region": np.random.choice(["Cairo","Alex","Giza"],3),
"Sales": np.random.randint(40000,80000,3),
})
df["Quarter"] = f"Q{q}"
quarters.append(df)
# Combine all quarters
annual = pd.concat(quarters, ignore_index=True)
print(f"Total rows: {len(annual)}")
print(f"Quarters: {annual['Quarter'].unique()}")
# Quarterly comparison
q_summary = annual.groupby("Quarter")["Sales"].agg(["sum","mean"]).round(0)
print("\nQuarterly Summary:")
print(q_summary)
# Verify: sum of quarters = annual total
print(f"\nAnnual total: {annual['Sales'].sum():,}")
print(f"Q-totals sum: {q_summary['sum'].sum():,}")
Creating Pivot Tables
python
import pandas as pd, numpy as np
np.random.seed(42)
df = pd.DataFrame({
"Region": np.random.choice(["Cairo","Alex","Giza"],100),
"Quarter": np.random.choice(["Q1","Q2","Q3","Q4"],100),
"Product": np.random.choice(["Laptop","Phone","Tablet"],100),
"Sales": np.random.randint(5000,80000,100),
})
# Basic pivot table
pivot = df.pivot_table(
values = "Sales",
index = "Region",
columns = "Quarter",
aggfunc = "sum",
margins = True, # Add totals
fill_value = 0,
)
print(pivot.round(0))
Output
Quarter Q1 Q2 Q3 Q4 All
Region
Alex 234000 198000 256000 287000 975000
Cairo 312000 345000 298000 321000 1276000
Giza 198000 212000 178000 245000 833000
All 744000 755000 732000 853000 3084000Multiple Values & Functions
python
# Multiple aggfunc
multi = df.pivot_table(
values = "Sales",
index = "Region",
columns = "Product",
aggfunc = ["sum","mean","count"],
).round(0)
# Cross-tabulation (counts)
ct = pd.crosstab(df["Region"], df["Quarter"],
values=df["Sales"], aggfunc="sum",
normalize="index" # Row percentages
) * 100
print(ct.round(1))
💡 Tip: Use margins=True to automatically add totals row and column. Use fill_value=0 to replace NaN with 0 in the pivot table.
🎯 TASKPractice: Pivot Tables
Create a comprehensive pivot analysis: revenue by region/quarter, market share by product, and identify best region-product combinations.
Starter code:
python
import pandas as pd, numpy as np
np.random.seed(42)
df = pd.DataFrame({
"Region": np.random.choice(["Cairo","Alex","Giza","Aswan"],200),
"Quarter": np.random.choice(["Q1","Q2","Q3","Q4"],200),
"Product": np.random.choice(["Laptop","Phone","Tablet","Monitor"],200),
"Sales": np.random.randint(3000,80000,200),
"Units": np.random.randint(1,10,200),
})
# Revenue pivot
rev_pivot = df.pivot_table(
values="Sales", index="Region",
columns="Quarter", aggfunc="sum",
margins=True, fill_value=0
)
print("Revenue by Region/Quarter:")
print(rev_pivot.round(0))
# Market share by product
product_share = (df.groupby("Product")["Sales"].sum() /
df["Sales"].sum() * 100).round(1)
print("\nProduct market share %:")
print(product_share.sort_values(ascending=False))
# Best combo
combo = df.groupby(["Region","Product"])["Sales"].sum()
print(f"\nBest combo: {combo.idxmax()} ({combo.max():,})")
Date Index & Resampling
python
import pandas as pd, numpy as np
np.random.seed(42)
# Create daily time series
dates = pd.date_range("2024-01-01","2024-12-31",freq="D")
df = pd.DataFrame({
"Sales": np.random.normal(50000,10000,len(dates)).round(0),
"Units": np.random.randint(1,20,len(dates)),
}, index=dates)
# Extract date components
df["Month"] = df.index.month
df["Quarter"] = df.index.quarter
df["DayName"] = df.index.day_name()
# Resample — aggregate by different frequencies
monthly = df["Sales"].resample("ME").sum()
weekly = df["Sales"].resample("W").mean()
quarterly = df["Sales"].resample("QE").agg(["sum","mean","count"])
print("Monthly Sales (first 3):")
print(monthly.head(3).round(0))
print("\nQuarterly:")
print(quarterly.round(0))
Output
Monthly Sales (first 3):
2024-01-31 1478234
2024-02-29 1389456
2024-03-31 1523891
Quarterly:
sum mean count
2024-03-31 4391581 48260.0 91
2024-06-30 4512345 49585.0 91🎯 TASKPractice: Time Series Basics
Analyze a year of daily sales: find seasonality patterns, best/worst day of week, month-over-month growth, and detect trend.
Starter code:
python
import pandas as pd, numpy as np
np.random.seed(42)
dates = pd.date_range("2023-01-01","2024-12-31",freq="D")
# Add trend + seasonality + noise
trend = np.linspace(40000,60000,len(dates))
seasonal = np.sin(np.linspace(0,4*np.pi,len(dates))) * 5000
noise = np.random.normal(0,3000,len(dates))
df = pd.DataFrame({"Sales": (trend+seasonal+noise).round(0)}, index=dates)
# Monthly totals and growth
monthly = df["Sales"].resample("ME").sum()
growth = monthly.pct_change() * 100
# Day of week analysis
df["DayName"] = df.index.day_name()
by_day = df.groupby("DayName")["Sales"].mean().sort_values(ascending=False)
# Best and worst months
print(f"Best month: {monthly.idxmax().strftime('%Y-%m')} ({monthly.max():,.0f})")
print(f"Worst month: {monthly.idxmin().strftime('%Y-%m')} ({monthly.min():,.0f})")
print(f"\nAvg growth: {growth.mean():.2f}% per month")
print(f"\nBest day of week: {by_day.index[0]} ({by_day.iloc[0]:,.0f})")
print(f"Worst day: {by_day.index[-1]} ({by_day.iloc[-1]:,.0f})")
Rolling Windows
python
import pandas as pd, numpy as np
monthly = pd.Series([45,52,48,61,58,72,65,59,71,78,82,95],
index=pd.date_range("2024-01",periods=12,freq="ME"),
name="Sales"
) * 1000
# Rolling mean (smoothing)
ma3 = monthly.rolling(window=3).mean() # 3-month moving avg
ma6 = monthly.rolling(window=6).mean() # 6-month moving avg
# Rolling statistics
roll_std = monthly.rolling(window=3).std()
roll_min = monthly.rolling(window=3).min()
roll_max = monthly.rolling(window=3).max()
result = pd.DataFrame({
"Sales": monthly,
"MA3": ma3,
"MA6": ma6,
"Std3": roll_std,
}).round(0)
print(result.head(8))
Output
Sales MA3 MA6 Std3
2024-01-31 45000 NaN NaN NaN
2024-02-29 52000 NaN NaN NaN
2024-03-31 48000 48333 NaN 3512
2024-04-30 61000 53667 NaN 6807
2024-05-31 58000 55667 NaN 6807
2024-06-30 72000 63667 56000 7211Expanding & Cumulative
python
# Expanding — includes all previous rows
cum_mean = monthly.expanding().mean()
cum_total = monthly.cumsum()
cum_max = monthly.cummax()
# Percentage change
pct = monthly.pct_change() * 100
result2 = pd.DataFrame({
"Sales": monthly,
"CumTotal": cum_total,
"RunningAvg":cum_mean,
"Growth%": pct,
}).round(0)
print(result2.tail(5))
💡 Tip: Rolling averages smooth out noise in time series. Use 3-period MA for short-term trend, 12-period for long-term.
🎯 TASKPractice: Window Functions
Apply window functions to daily sales: calculate 7-day and 30-day moving averages, detect days where sales spike > 2 std above rolling mean, and calculate cumulative revenue milestones.
Starter code:
python
import pandas as pd, numpy as np
np.random.seed(42)
dates = pd.date_range("2024-01-01","2024-06-30",freq="D")
sales = pd.Series(
np.random.normal(50000,8000,len(dates)).round(0),
index=dates, name="Sales"
)
# Moving averages
ma7 = sales.rolling(7).mean()
ma30 = sales.rolling(30).mean()
# Rolling std for spike detection
roll_std = sales.rolling(30).std()
roll_mean = sales.rolling(30).mean()
spikes = sales[sales > roll_mean + 2*roll_std]
# Cumulative revenue
cum_rev = sales.cumsum()
milestones = [1e6, 2e6, 3e6]
for m in milestones:
date = cum_rev[cum_rev >= m].index[0] if (cum_rev >= m).any() else None
if date: print(f"Reached {m/1e6:.0f}M on: {date.date()}")
print(f"\nSpike days: {len(spikes)}")
print(f"Avg spike amount: {spikes.mean():,.0f}")
print(f"\nFinal 7-day MA: {ma7.iloc[-1]:,.0f}")
Correlation Analysis
python
import pandas as pd, numpy as np
np.random.seed(42)
n = 100
marketing = np.random.normal(5000, 1000, n)
sales = marketing * 8 + np.random.normal(0, 5000, n) # Correlated
df = pd.DataFrame({
"Marketing": marketing.round(0),
"Sales": sales.round(0),
"Staff": np.random.randint(5, 30, n),
"Season": np.random.choice([1,2,3,4], n),
})
# Pearson correlation
print("Correlation matrix:")
print(df.corr().round(3))
# Spearman (rank-based, robust to outliers)
print("\nSpearman correlation:")
print(df.corr(method="spearman").round(3))
Output
Correlation matrix:
Marketing Sales Staff Season
Marketing 1.000 0.845 0.012 -0.023
Sales 0.845 1.000 0.089 -0.041
Staff 0.012 0.089 1.000 0.054
Season -0.023 -0.041 0.054 1.000Simple Linear Regression
python
import numpy as np
# Manual linear regression
X = df["Marketing"].values
y = df["Sales"].values
# OLS: y = a + bX
b = np.cov(X, y)[0,1] / np.var(X)
a = y.mean() - b * X.mean()
# Predictions
y_pred = a + b * X
residuals = y - y_pred
r_squared = 1 - (residuals**2).sum() / ((y - y.mean())**2).sum()
print(f"Intercept (a): {a:,.0f}")
print(f"Slope (b): {b:.2f}")
print(f"R²: {r_squared:.3f}")
print(f"\nFor 6000 marketing: {a + b*6000:,.0f} predicted sales")
🎯 TASKPractice: Correlation & Regression
Perform correlation analysis on a multi-variable dataset. Find the variables most correlated with sales, build a simple regression, and interpret results.
Starter code:
python
import pandas as pd, numpy as np
np.random.seed(42)
n = 150
# Generate correlated data
marketing = np.random.normal(5000,1000,n)
staff = np.random.randint(5,30,n)
season_eff = np.random.choice([0.8,1.0,1.2,1.1],n)
sales = (marketing*7 + staff*500 + np.random.normal(0,3000,n)) * season_eff
df = pd.DataFrame({
"Sales": sales.round(0),
"Marketing": marketing.round(0),
"Staff": staff,
"SeasonEff": season_eff,
"Temperature": np.random.normal(25,5,n).round(1),
})
# Correlation with Sales
corr_with_sales = df.corr()["Sales"].drop("Sales").sort_values(ascending=False)
print("Correlation with Sales:")
print(corr_with_sales.round(3))
# Simple regression: Sales ~ Marketing
X = df["Marketing"].values
y = df["Sales"].values
b = np.cov(X,y)[0,1]/np.var(X)
a = y.mean()-b*X.mean()
y_pred = a+b*X
r2 = 1-((y-y_pred)**2).sum()/((y-y.mean())**2).sum()
print(f"\nRegression: Sales = {a:,.0f} + {b:.2f} × Marketing")
print(f"R² = {r2:.3f} ({r2*100:.1f}% variance explained)")
Random Sampling
python
import pandas as pd, numpy as np
np.random.seed(42)
# Large population
population = pd.DataFrame({
"Sales": np.random.normal(50000,15000,10000).round(0),
"Region": np.random.choice(["Cairo","Alex","Giza"],10000),
})
# Simple random sample
sample = population.sample(n=100, random_state=42)
print(f"Population mean: {population['Sales'].mean():,.0f}")
print(f"Sample mean: {sample['Sales'].mean():,.0f}")
# Stratified sample (proportional by region)
stratified = population.groupby("Region", group_keys=False).apply(
lambda x: x.sample(frac=0.01, random_state=42)
)
print(f"\nStratified sample: {len(stratified)} rows")
print(stratified["Region"].value_counts())
Output
Population mean: 50,012
Sample mean: 49,847
Stratified sample: 100 rows
Region
Cairo 34
Alex 33
Giza 33Hypothesis Testing (t-test)
python
import numpy as np
from scipy import stats
# Are Cairo sales significantly higher than Alex?
cairo = population[population["Region"]=="Cairo"]["Sales"]
alex = population[population["Region"]=="Alex"]["Sales"]
t_stat, p_value = stats.ttest_ind(cairo, alex)
print(f"Cairo mean: {cairo.mean():,.0f}")
print(f"Alex mean: {alex.mean():,.0f}")
print(f"t-statistic: {t_stat:.3f}")
print(f"p-value: {p_value:.4f}")
print(f"Significant (p<0.05): {p_value < 0.05}")
💡 Tip: p-value < 0.05 means we reject the null hypothesis — the difference is statistically significant (not due to random chance).
🎯 TASKPractice: Sampling & Hypothesis Testing
Compare sales between two marketing campaigns using hypothesis testing. Determine if the difference is statistically significant.
Starter code:
python
import numpy as np, pandas as pd
from scipy import stats
np.random.seed(42)
# Two campaigns
campaign_A = np.random.normal(52000, 8000, 50) # New campaign
campaign_B = np.random.normal(48000, 9000, 50) # Old campaign
# Descriptive stats
print("Campaign A:")
print(f" Mean: {campaign_A.mean():,.0f} Std: {campaign_A.std():,.0f}")
print("Campaign B:")
print(f" Mean: {campaign_B.mean():,.0f} Std: {campaign_B.std():,.0f}")
print(f" Difference: {campaign_A.mean()-campaign_B.mean():,.0f}")
# Independent t-test
t_stat, p_val = stats.ttest_ind(campaign_A, campaign_B)
print(f"\nt-statistic: {t_stat:.3f}")
print(f"p-value: {p_val:.4f}")
if p_val < 0.05:
print("✅ SIGNIFICANT: Campaign A performs better (p < 0.05)")
else:
print("❌ NOT significant: Could be random variation (p >= 0.05)")
# Effect size (Cohen's d)
pooled_std = np.sqrt((campaign_A.std()**2 + campaign_B.std()**2) / 2)
cohens_d = (campaign_A.mean() - campaign_B.mean()) / pooled_std
print(f"Cohen's d: {cohens_d:.3f} ({'large' if abs(cohens_d)>0.8 else 'medium' if abs(cohens_d)>0.5 else 'small'} effect)")
Structuring an Analysis Report
python
import pandas as pd, numpy as np
np.random.seed(42)
def sales_report(df, title="Sales Report"):
"""Generate a complete sales analysis report."""
sep = "="*50
print(f"\n{sep}")
print(f" {title}")
print(sep)
# 1. Overview
print(f"\n📊 OVERVIEW")
print(f" Period: {df.index.min().date()} to {df.index.max().date()}")
print(f" Total: {df['Sales'].sum():>15,.0f}")
print(f" Daily Avg: {df['Sales'].mean():>15,.0f}")
print(f" Best Day: {df['Sales'].max():>15,.0f} ({df['Sales'].idxmax().date()})")
# 2. Trend
monthly = df["Sales"].resample("ME").sum()
growth = monthly.pct_change().mean() * 100
print(f"\n📈 TREND")
print(f" Monthly growth (avg): {growth:+.1f}%")
# 3. Regional breakdown
if "Region" in df.columns:
print(f"\n🗺️ BY REGION")
by_region = df.groupby("Region")["Sales"].agg(["sum","mean"]).round(0)
by_region["Share%"] = (by_region["sum"]/by_region["sum"].sum()*100).round(1)
print(by_region.sort_values("sum",ascending=False).to_string())
print(f"\n{sep}\n")
return monthly
💡 Tip: A good report has: Overview → Trend → Breakdown → Insights → Recommendations. Always answer the 'so what' question.
🎯 TASKPractice: Building Analysis Reports
Build a full quarterly business report with: executive summary, KPI dashboard, regional analysis, product performance, and growth analysis.
Starter code:
python
import pandas as pd, numpy as np
np.random.seed(42)
# Generate dataset
dates = pd.date_range("2024-01-01","2024-12-31",freq="D")
regions = np.random.choice(["Cairo","Alex","Giza","Aswan"],len(dates))
products = np.random.choice(["Laptop","Phone","Tablet"],len(dates))
sales = np.random.normal(50000,15000,len(dates)).round(0)
df = pd.DataFrame({"Sales":sales,"Region":regions,"Product":products},index=dates)
print("=" * 55)
print(" ANNUAL BUSINESS REPORT — 2024")
print("=" * 55)
# KPIs
print(f"\n{'KPI':<25} {'Value':>15}")
print("-"*40)
print(f"{'Total Revenue':<25} {df['Sales'].sum():>15,.0f}")
print(f"{'Daily Average':<25} {df['Sales'].mean():>15,.0f}")
print(f"{'Best Month Revenue':<25}", end="")
monthly=df["Sales"].resample("ME").sum()
print(f"{monthly.max():>15,.0f}")
print(f"{'YoY Growth Est.':<25} {'N/A - 1 year':>15}")
# Regional breakdown
print(f"\n{'REGIONAL PERFORMANCE':=^40}")
reg=df.groupby("Region")["Sales"].agg(Total="sum",Daily="mean").round(0)
reg["Share%"]=(reg["Total"]/reg["Total"].sum()*100).round(1)
print(reg.sort_values("Total",ascending=False).to_string())
# Product mix
print(f"\n{'PRODUCT MIX':=^40}")
prod=df.groupby("Product")["Sales"].sum()
for p,v in prod.sort_values(ascending=False).items():
share=v/prod.sum()*100
bar="█"*int(share/5)
print(f" {p:<10} {v:>10,.0f} ({share:.1f}%) {bar}")