What data science is, and how this course works

Why data science exists

Every shop, hospital, bank and app collects data: bills, visits, payments, clicks. On its own, that pile of data does not help anyone. Data science is the work of turning data into answers that help someone decide what to do. "Which items should the kirana shop stock more of before Diwali?" is a question. The sales register is the data. The answer, with evidence, is what data science produces.

The data workflow

Most data projects follow the same six steps. Real projects loop back often: exploring the data usually shows a cleaning problem you missed, and the answer usually raises a new question.

1. Question: write down exactly what you want to know and which decision it will help. A good question is specific and measurable.
2. Collect: find or gather the data that can answer it: a sales register, a spreadsheet, a database, a government data portal, a survey.
3. Clean: fix the mess. Remove duplicate rows, fill or drop blank values, make "Pune", "pune " and "PUNE" the same city, turn text like "₹1,250" into the number 1250.
4. Explore: count, total, average and chart the data to understand it. Most useful answers are found in this step.
5. Model: when the question needs a prediction ("how many packets of milk will sell tomorrow?"), build a machine learning model. Many questions never need this step.
6. Communicate: explain the answer with a short summary and a clear chart, including how sure you are and what the data cannot tell you.

Who does what

Job titles overlap and differ between companies, but this is the usual split. Module 12 covers careers, skills and salaries in detail.

Structured and unstructured data

Structured data fits neatly in a table: every row is one record (one sale, one patient) and every column is one property (date, amount, city). Spreadsheets and SQL databases hold structured data. Unstructured data has no fixed table shape: customer reviews, WhatsApp messages, photos of bills, call recordings. In between sits semi-structured data such as JSON, which has labels but not a fixed table shape. This course starts with structured data, and Modules 9 and 10 show how neural networks and language models handle text and images.

How this course works

The course has 12 modules in three levels. At about 1 hour a day, five days a week, it takes about 20 weeks.

• Weeks 1–7, Beginner (Modules 1–4): Python for data, statistics, probability, and cleaning, exploring and charting data.
• Weeks 8–14, Intermediate (Modules 5–8): the machine learning workflow, regression and classification, evaluating models, trees and ensembles, clustering and feature engineering.
• Weeks 15–20, Expert (Modules 9–12): neural networks, NLP and large language models, building with generative AI, MLOps, ethics and careers.

The tools

Python is the main language of data work because it is easy to read and has excellent free libraries. A library is a package of ready-made code you import. Jupyter is a notebook: a page where you run code in small cells and see results, tables and charts right below each cell. pandas works with tables, NumPy does fast maths on arrays of numbers, matplotlib draws charts and scikit-learn builds machine learning models.

Here is the plan for practice. The coding tasks in this course run in your browser using only Python's built-in standard library, so you learn what the big libraries do underneath. The library versions are shown next to them, with their output, so you will recognise them at work.

sales = [120, 450, 80, 300, 950]   # one day's bills at a kirana shop, in rupees

print(sum(sales))                # total
print(len(sales))                # number of bills
print(sum(sales) / len(sales))   # average bill
print(max(sales))                # biggest bill

import pandas as pd

df = pd.DataFrame({"item": ["milk", "bread", "rice"], "amount": [120, 45, 300]})
print(df["amount"].sum())   # 465

Common mistakes

• Starting with the data instead of a question. Fix: write the question and the decision it supports in one sentence before you open any file.
• Asking a vague question such as "analyse the sales". Fix: make it measurable: "Which 5 items earned the most in October, and did that change from September?"
• Skipping cleaning. Fix: always count blanks, duplicates and strange values before you calculate anything.
• Hiding uncertainty. Fix: say how much data you used and what it cannot show, for example "only 3 weeks of data, so festival weeks are not included".