🚀 Day-Before Cheatsheet — Rapid Revision

Everything in One Page — Read This the Night Before

---

SQL — Top 20 Patterns

#	Pattern	Syntax
1	Select specific columns	SELECT col1, col2 FROM table
2	Filter rows	WHERE col = 'value'
3	Multiple conditions	WHERE a = 1 AND (b = 2 OR c = 3)
4	Pattern match	WHERE name LIKE 'A%'
5	Null check	WHERE col IS NULL / IS NOT NULL
6	Sort results	ORDER BY col DESC
7	Limit rows	LIMIT 10 / TOP 10
8	Aggregate	SELECT dept, COUNT(*), AVG(sal) FROM t GROUP BY dept
9	Filter groups	GROUP BY dept HAVING COUNT(*) > 5
10	Inner Join	FROM a JOIN b ON a.id = b.id
11	Left Join	FROM a LEFT JOIN b ON a.id = b.id
12	Self Join	FROM emp e JOIN emp m ON e.mgr_id = m.id
13	Subquery in WHERE	WHERE sal > (SELECT AVG(sal) FROM emp)
14	Subquery in FROM	FROM (SELECT ... ) AS sub
15	CASE statement	CASE WHEN x > 0 THEN 'Yes' ELSE 'No' END
16	Window — RANK	RANK() OVER (PARTITION BY dept ORDER BY sal DESC)
17	Window — Running Total	SUM(amt) OVER (ORDER BY date)
18	Window — LAG/LEAD	LAG(amt) OVER (ORDER BY date)
19	CTE	WITH cte AS (SELECT ...) SELECT * FROM cte
20	Delete duplicates	DELETE FROM t WHERE id NOT IN (SELECT MIN(id) FROM t GROUP BY col)

---

DAX — Top 10 Formulas

#	Formula	Use
1	SUM(Sales[Revenue])	Basic total
2	COUNTROWS(Sales)	Count rows
3	DIVIDE(A, B, 0)	Safe division
4	DISTINCTCOUNT(Sales[CustID])	Unique count
5	CALCULATE(SUM(...), filter)	Change filter context
6	TOTALYTD(SUM(...), Date[Date])	Year-to-date
7	SAMEPERIODLASTYEAR(Date[Date])	Previous year comparison
8	ALL(Table)	Remove all filters
9	SUMX(table, expr)	Row-by-row calculation
10	CALCULATE + ALL → % of Total	Percentage analysis

---

Tableau — 5 Must-Know Concepts

1. Dimensions vs Measures: Blue = categorical (creates headers). Green = numerical (gets aggregated).
2. Context Filter: Creates a temp table BEFORE other filters. Use for "Top N within a group."
3. LOD — FIXED: {FIXED [Region] : SUM([Sales])} — calculates at specified level, ignores the view.
4. Filter Order: Extract → Data Source → Context → Dimension → Measure → Table Calc.
5. Live vs Extract: Extract = faster (local copy, .hyper file). Live = real-time but slower. Use Extract 90% of the time.

---

Statistics — 5 Core Concepts

1. Mean vs Median: Mean = sum/count (sensitive to outliers). Median = middle value (robust). Use Median for skewed data.
2. 68-95-99.7 Rule: In a normal distribution, 68% within ±1σ, 95% within ±2σ, 99.7% within ±3σ.
3. p-value: If p < 0.05 → result is statistically significant → reject H₀.
4. Type I vs II: Type I = false positive (wrongly rejected H₀). Type II = false negative (missed real effect).
5. Correlation ≠ Causation: r measures linear relationship. To prove causation → need controlled experiment (A/B test).

---

Python — 5 One-Liners

# 1. Read CSV and show shape
df = pd.read_csv('data.csv'); print(df.shape)

# 2. Missing values per column
df.isnull().sum()

# 3. Group by and aggregate
df.groupby('city')['revenue'].agg(['sum', 'mean', 'count'])

# 4. Top 10 rows by a column
df.nlargest(10, 'revenue')

# 5. Quick correlation matrix
df[['revenue', 'profit', 'quantity']].corr()

---

Excel — 5 Key Formulas

Formula	What It Does
=INDEX(C:C, MATCH("X", A:A, 0))	Better VLOOKUP — looks any direction
=SUMIFS(D:D, B:B, "Delhi", C:C, ">1000")	Conditional sum
=COUNTIFS(A:A, "Delhi", B:B, ">50000")	Conditional count
=IFERROR(VLOOKUP(...), "Not Found")	Handle lookup errors
=TEXT(A1, "MMM-YY")	Format dates

---

Interview Mantras — Say These Lines

Topic	Power Line
Outliers	"I'd use Median, not Mean, because Mean is sensitive to outliers."
A/B Test	"I'd ensure sufficient sample size with a power analysis before starting."
Accuracy Paradox	"98% accuracy is meaningless with 2% fraud — I'd use F1 Score and AUC."
Context Filter	"It creates a temporary table — subsequent filters work within that subset."
CALCULATE	"It modifies the filter context of a DAX expression."
Correlation	"Correlation suggests a relationship but doesn't prove causation."
Decision Tree	"Easy to interpret but prone to overfitting. Random Forest fixes this via ensemble averaging."
Star Schema	"Fact table = events. Dimensions = context. Filters flow dim → fact."