A data science interview at a mid-size startup once opened with this: "Without Googling, what's the difference between variance and standard deviation, and why does it matter for model selection?" Half the room went quiet. That's exactly what a solid data science cheat sheet prevents — not because you'll bring it to an interview, but because drilling the reference material locks the concepts in place.
This data science cheat sheet covers the core areas you'll hit repeatedly: statistics fundamentals, Python libraries, machine learning algorithms, SQL patterns, and visualization principles. Each section is dense by design — this is reference material, not a tutorial.
Statistics & Probability: The Data Science Cheat Sheet Core
Statistics is where most newcomers either build a real foundation or fake their way through until it catches up with them. These are the concepts that appear constantly.
Descriptive Statistics
- Mean: Sum of values / count. Sensitive to outliers.
- Median: Middle value when sorted. Robust to outliers — prefer for skewed distributions.
- Mode: Most frequent value. Useful for categorical data.
- Variance (σ²): Average squared deviation from the mean.
Σ(xᵢ − μ)² / N - Standard Deviation (σ): Square root of variance. Same units as the original data — more interpretable.
- IQR: Q3 − Q1. The middle 50% spread. Outlier threshold: values outside Q1 − 1.5×IQR or Q3 + 1.5×IQR.
Key Distributions
- Normal: Bell curve. 68-95-99.7 rule (1σ, 2σ, 3σ from mean).
- Binomial: Count of successes in N trials with probability p.
- Poisson: Events per fixed interval when events are rare and independent.
- Uniform: Equal probability across range. Common in simulation.
Hypothesis Testing
- p-value: Probability of observing results at least as extreme as your data, assuming H₀ is true. A p-value of 0.03 does NOT mean "97% chance the effect is real."
- Type I Error (α): Rejecting a true null hypothesis. False positive. Usually set at 0.05.
- Type II Error (β): Failing to reject a false null hypothesis. False negative.
- Statistical Power: 1 − β. The probability of detecting a real effect.
- t-test: Compare means of two groups. Use Welch's when variances differ.
- Chi-squared test: Test independence between categorical variables.
- ANOVA: Compare means across 3+ groups simultaneously.
Python Libraries: Essential Data Science Cheat Sheet Reference
Python dominates data science workflows. These are the libraries and their most-used patterns.
NumPy
import numpy as np
arr = np.array([1, 2, 3])
arr.mean() # mean
arr.std() # standard deviation
arr.reshape(3, 1) # reshape
np.dot(A, B) # matrix multiplication
np.linspace(0, 1, 100) # 100 evenly spaced pointsPandas
import pandas as pd
df = pd.read_csv('data.csv')
df.info() # dtypes + null counts
df.describe() # descriptive stats
df.isnull().sum() # missing per column
df.groupby('col').agg({'val': 'mean'})
df.merge(df2, on='id', how='left')
df['col'].value_counts()
df.drop_duplicates(subset=['id'])
df.fillna(df.median()) # impute with medianScikit-learn Pipeline Pattern
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42)
pipe = Pipeline([
('scaler', StandardScaler()),
('model', RandomForestClassifier(n_estimators=100))
])
pipe.fit(X_train, y_train)
print(classification_report(y_test, pipe.predict(X_test)))Machine Learning Algorithms: Quick Reference
Knowing when to use which algorithm separates practitioners from people who just run model.fit() and hope.
Algorithm Selection Guide
| Problem | Algorithm | When to Use |
|---|---|---|
| Binary Classification | Logistic Regression | Baseline; interpretable coefficients |
| Binary / Multi Classification | Random Forest | Non-linear, handles mixed types, robust to outliers |
| Classification / Regression | XGBoost / LightGBM | Tabular data competitions; often best out-of-box |
| Regression | Linear Regression | Baseline; continuous target; interpretable |
| Clustering | K-Means | Known k; spherical clusters; fast |
| Clustering | DBSCAN | Unknown k; arbitrary shapes; finds outliers |
| Dimensionality Reduction | PCA | Remove correlated features; speed up training |
| NLP | TF-IDF + LogReg | Text classification baseline |
| NLP | BERT fine-tune | When accuracy matters more than speed |
Key Metrics by Task
- Classification: Accuracy, Precision, Recall, F1, AUC-ROC. Use F1 when classes are imbalanced. Use AUC when threshold is uncertain.
- Regression: MAE (interpretable), RMSE (penalizes large errors), R² (proportion of variance explained).
- Clustering: Silhouette score, Inertia (within-cluster sum of squares).
Overfitting Checklist
- Training accuracy high, validation accuracy low → overfitting
- Fixes: more data, regularization (L1/L2), dropout (neural nets), reduce model complexity, cross-validation
- L1 (Lasso): drives some coefficients to zero → feature selection
- L2 (Ridge): shrinks all coefficients → handles multicollinearity
SQL Patterns Every Data Scientist Uses
Most real-world data lives in databases, not CSVs. These patterns cover 90% of what you'll write day-to-day.
-- Window function: rank users by revenue within each region
SELECT user_id, region, revenue,
RANK() OVER (PARTITION BY region ORDER BY revenue DESC) AS rank
FROM orders;
-- Rolling 7-day average
SELECT date, revenue,
AVG(revenue) OVER (ORDER BY date ROWS BETWEEN 6 PRECEDING AND CURRENT ROW) AS rolling_7d
FROM daily_sales;
-- Find duplicates
SELECT email, COUNT(*) as cnt
FROM users
GROUP BY email
HAVING COUNT(*) > 1;
-- Cohort retention skeleton
SELECT cohort_month,
SUM(CASE WHEN months_since = 0 THEN 1 END) AS month_0,
SUM(CASE WHEN months_since = 1 THEN 1 END) AS month_1
FROM cohort_data
GROUP BY cohort_month;Top Courses to Build These Skills
A cheat sheet shows you the map. These courses give you the territory — actual practice with real datasets.
Introduction to Data Analytics
Covers the full data analysis workflow from problem framing to visualization. A strong starting point if the statistics section above had more unknowns than knowns.
Executive Data Science Specialization
Four-course series covering the data science process end-to-end — useful for understanding how all the cheat-sheet pieces fit together in real projects with stakeholders and timelines.
Applied Plotting, Charting & Data Representation in Python
Goes deep on matplotlib and data visualization principles — the practical complement to the NumPy/Pandas section above. Teaches you to communicate findings, not just produce them.
COVID-19 Data Analysis Using Python
Works through a real-world messy dataset using Pandas, Matplotlib, and Seaborn. Better than toy datasets for building pattern recognition on data cleaning problems.
Database Design and Basic SQL in PostgreSQL
Covers the SQL fundamentals behind the window functions and GROUP BY patterns in this cheat sheet. PostgreSQL syntax transfers cleanly to BigQuery, Redshift, and Snowflake.
Introduction to Data Analysis using Microsoft Excel
If you work in a non-engineering environment where Python isn't the default, Excel pivot tables and formulas remain genuinely useful. This course covers them systematically.
FAQ
What should a data science cheat sheet include?
At minimum: statistics fundamentals (distributions, hypothesis testing, key formulas), Python library syntax for NumPy/Pandas/Scikit-learn, an algorithm selection guide, key evaluation metrics per task type, and SQL patterns for data extraction. Add visualization libraries and feature engineering techniques once you have the core down.
Is Python or R better for data science?
Python dominates in production environments, ML engineering, and most job postings. R has deeper statistical libraries and is preferred in academic research and clinical settings. If you're choosing one to start: Python. If your target industry is bioinformatics, pharma, or academic research: R is worth learning alongside Python.
How do I choose between machine learning algorithms?
Start with a simple baseline (logistic regression for classification, linear regression for regression). Beat the baseline before reaching for complexity. For tabular data, gradient boosting (XGBoost/LightGBM) is the empirically strongest default. Deep learning is worth the overhead only when you have large datasets and unstructured inputs (images, text, audio).
What's the difference between precision and recall?
Precision: of all the cases your model flagged positive, what fraction actually were? Recall: of all actual positives, what fraction did your model catch? High precision matters when false positives are costly (spam filter flagging real emails). High recall matters when false negatives are costly (cancer screening missing cases). F1 score is the harmonic mean of both.
Do I need to memorize all of this?
No — but you need to recognize these concepts when you see them and know when to apply them. The value of a cheat sheet is repeated exposure until the patterns become automatic, not memorization before an interview. Work through at least one project in each area.
What SQL dialect should I learn first?
Standard SQL (ANSI SQL) transfers across PostgreSQL, MySQL, BigQuery, Redshift, and Snowflake with minor syntax differences. PostgreSQL is the best learning environment — it's free, strict about types (which teaches good habits), and has full window function support. The window function and aggregation patterns above are valid in all major dialects.
Bottom Line
The most useful thing to do with a data science cheat sheet is work through it actively rather than read it once and save the tab. Pick any section above where you had to slow down — that's where to spend your practice time next. The statistics section trips up most beginners (especially hypothesis testing). The SQL window functions trip up people who learned data science through Python and skipped databases. The algorithm selection table is useless until you've actually implemented 3-4 of those algorithms and seen where they fail.
If you're earlier in the journey, Introduction to Data Analytics and the PostgreSQL SQL course cover the foundational material in this cheat sheet with hands-on practice. If you're post-foundation and want to level up the statistics and ML sections, the Executive Data Science Specialization connects these concepts to real decision-making workflows.