Python for Data Science: A Beginner's Roadmap

Why Python Dominates Data Science

Data science runs on Python. Not partly — overwhelmingly. In Kaggle's 2024 annual survey, over 85% of data practitioners use Python as their primary language. The reasons are practical: NumPy and pandas make data manipulation fast, matplotlib and seaborn make visualisation easy, and scikit-learn and PyTorch have become the default tools for machine learning.

If you want to work with data professionally, Python is not a choice — it is a prerequisite.

---

The Core Stack (in Learning Order)

1. Core Python First

Before touching any data library, you need to be comfortable with vanilla Python:

• Lists, dictionaries, and tuples
• List comprehensions
• Functions (including lambda)
• File I/O (open(), reading CSV with the built-in csv module)
• Basic OOP (enough to understand class-based APIs)

If you cannot write a function that reads a CSV file, counts the occurrences of each value in a column, and returns the top 5, spend more time on core Python first.

2. NumPy — Fast Numerical Computation

NumPy introduces the ndarray — a multi-dimensional array that is far faster than Python lists for numerical operations because it is implemented in C.

import numpy as np

# Create a 3x3 matrix
matrix = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]])

# Vectorised operations — no loops needed
print(matrix * 2)
print(matrix.mean(axis=0))  # Mean of each column
print(np.sqrt(matrix))       # Element-wise square root

You do not need to master NumPy before moving on, but you need to understand arrays, shapes, axes, and broadcasting.

3. pandas — Data Manipulation

pandas is the workhorse of data science in Python. Its two core objects — Series (1D) and DataFrame (2D) — wrap NumPy arrays and add labels, indexes, and a powerful API for loading, cleaning, transforming, and analysing data.

import pandas as pd

df = pd.read_csv("sales.csv")

# Inspect
print(df.head())
print(df.dtypes)
print(df.describe())

# Filter, group, aggregate
monthly = (
    df[df["region"] == "North"]
    .groupby("month")["revenue"]
    .sum()
    .reset_index()
)
print(monthly)

Focus on: loading data (read_csv, read_json), selecting rows/columns, filtering, groupby, merge, handling missing values (dropna, fillna).

4. Matplotlib and Seaborn — Visualisation

Data you cannot explain is data nobody acts on. matplotlib gives you low-level control; seaborn sits on top of it and produces statistical charts in one line.

import matplotlib.pyplot as plt
import seaborn as sns

# Distribution plot
sns.histplot(df["revenue"], bins=30, kde=True)
plt.title("Revenue Distribution")
plt.show()

# Correlation heatmap
sns.heatmap(df.corr(numeric_only=True), annot=True, fmt=".2f")
plt.show()

5. scikit-learn — Machine Learning

scikit-learn provides a consistent API for most classical ML models. You call .fit(), .predict(), and .score() on almost everything, which makes switching between models trivially easy.

from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

model = RandomForestClassifier(n_estimators=100, random_state=42)
model.fit(X_train, y_train)

preds = model.predict(X_test)
print(f"Accuracy: {accuracy_score(y_test, preds):.2%}")

---

A Realistic Data Science Workflow

Most real data projects follow this pattern:

1. Define the question — what decision will this analysis inform?
2. Get the data — CSV, database query, API, web scraping
3. Explore — df.describe(), distributions, nulls, outliers
4. Clean — handle missing values, fix types, remove duplicates
5. Engineer features — create new columns from existing ones
6. Model — fit a baseline model, then iterate
7. Evaluate — cross-validation, confusion matrix, business metrics
8. Communicate — a Jupyter notebook, a dashboard, or a slide deck

The step most beginners skip is step 1. Jumping straight to code without a clear question produces analysis nobody uses.

---

The Skill Most Tutorials Skip: Communication

You can be technically excellent and professionally invisible if you cannot explain what you found.

Get comfortable with:

• Writing Jupyter notebooks that tell a story (not just code cells)
• Matplotlib/seaborn charts with proper titles, labels, and captions
• Plain-language summaries of statistical findings

---

Your First Project

Do not start with MNIST. Start with a dataset about something you actually care about — football stats, financial markets, music, whatever — and answer a single question that is interesting to you.

The Kaggle datasets library has hundreds of clean, real-world CSV files. Download one, open a Jupyter notebook, and work through the five-step workflow above.

---

Start with Python Fundamentals

Data science is applied Python. Before you touch pandas, make sure your core Python is solid. Python Foundry's certification course covers everything from variables to OOP with in-browser exercises — a strong foundation before you move into the data stack.