Python for Data Analysis in 2026: A 10-Week Beginner Roadmap With Weekly Milestones

You have been given a 48,000 row CSV of claims data, the file opens in Excel but freezes on every pivot, and the answer your manager wants is a single number by city. That is the moment most people in Indian operations and finance teams first hear that they should "just use Python". The advice is right. The problem is that the internet will then hand you a Django tutorial, a linear algebra course and a 40 hour video that never opens a real file.

This is the shortest honest path from that stuck spreadsheet to a working analysis, at about 8 spare hours a week. The running example is a claims processing executive in Pune, two years in, comfortable in Excel, no programming background.

What Python for Data Analysis Actually Replaces

Python for data analysis is a narrow slice of Python. You are not learning to build software. You are learning to describe a table operation in code so it runs the same way on 48,000 rows as on 48, and so next month's file takes 20 seconds instead of an afternoon.

Four things move out of Excel and into code: loading messy files, cleaning columns, grouping and joining, and producing a chart. Everything else sits on top of those four.

The four skills employers actually screen for

Analyst job descriptions in India cluster tightly around SQL, Excel, one BI tool, and enough statistics to know when a number is meaningless. Python usually appears as a strong fifth: valued, increasingly expected, rarely the thing that rejects you on its own. That ordering is why the plan below gives SQL two full weeks rather than a footnote, and why 360DT's Data Analyst course teaches the four together with Python alongside.

Also read: How to Become a Data Analyst Without Experience in India for the wider career switch view around this skill plan.

The 10-Week Python for Data Analysis Roadmap

Here is the plan in full. Each block ends in something you can show another person, because a milestone you cannot demonstrate is not a milestone.

Where the hours actually go

Beginners over-invest in syntax and under-invest in cleaning. This is what ten weeks looks like when the real work sets the timetable.

Weeks 1 and 2: Install Python Once, Properly

Most people lose their first fortnight to environments. Anaconda, then a second Python from the Microsoft Store, then pip puts pandas in the wrong one, and by the time the import error appears they have concluded they are bad at this.

Use uv, the Rust based installer, and use only uv. It installs packages far faster than pip and manages the Python version too, so there is exactly one place where things live. The trade-off I will accept: uv is newer than conda and a few scientific packages still have rough edges. For a laptop that needs pandas, DuckDB and Jupyter, it removes the whole class of problem that makes beginners quit.

# one-time setup, works on Windows PowerShell, macOS and Linux
uv init --bare claims-analysis
cd claims-analysis
uv add pandas duckdb matplotlib
uv add --dev jupyterlab

# start the notebook server
uv run jupyter lab

Python 3.14.7 is the current stable release on python.org as of 5 August 2026, and pandas 3.0 requires Python 3.11 or newer, so anything from 3.11 up is safe. If a tutorial tells you to install Python 3.8 because "it is more compatible", the tutorial is old.

What usually goes wrong here

You will type pip install pandas out of habit, it will land in some other Python, and the notebook will still say ModuleNotFoundError. Stop debugging and check the interpreter: run import sys; print(sys.executable) in a cell. If that path is not inside your project's .venv folder, no amount of reinstalling pandas will fix it.

Weeks 3 to 5: A Pandas Tutorial for Beginners That Ends in a Real Answer

A pandas tutorial for beginners usually stops at printing a DataFrame. Go further in one sitting: load a file, fix its types, group it, produce the number someone asked for. Here is the whole loop, with output you can reproduce.

The input file, orders.csv:

order_id,city,channel,amount
1001,Pune,Online,1200
1002,Pune,Retail,800
1003,Mumbai,Online,2500
1004,Mumbai,Online,1500
1005,Mumbai,Retail,1000
1006,Pune,Online,400
1007,Chennai,Retail,900
1008,Chennai,Online,1100

import pandas as pd

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

# 1. headers arrive with stray spaces more often than you would like
df.columns = df.columns.str.strip()

# 2. force the money column to be numeric; bad cells become NaN, not crashes
df["amount"] = pd.to_numeric(df["amount"], errors="coerce")

# 3. drop rows that have no usable amount
df = df[df["amount"].notna()]

# 4. the actual question: revenue and average order value by city
summary = (df.groupby("city")["amount"]
             .agg(["sum", "mean", "count"])
             .sort_values("sum", ascending=False)
             .round(2))

print(summary)

Running that prints:

            sum     mean  count
city
Mumbai     5000  1666.67      3
Pune       2400   800.00      3
Chennai    2000  1000.00      2

Nine lines, and you have replaced a pivot table that needed rebuilding every month. Change the file, re-run the cell, the answer updates. That is the entire pitch for Python for data analysis, and seeing it work on your own data solves the motivation problem.

pandas 3.0 changed the rules under the tutorials

pandas 3.0.0 landed on 21 January 2026 and the current stable version is 3.0.5, released on 22 July 2026 according to the official pandas release notes. Two changes matter to you on day one. Copy-on-Write is now the only mode, so any indexing operation behaves as a copy, which means chained assignment such as df[df["amount"] > 0]["flag"] = 1 silently fails to change your DataFrame. Use df.loc[df["amount"] > 0, "flag"] = 1 instead, always.

The second change: string columns are now inferred as a dedicated str dtype backed by Apache Arrow rather than the old object dtype. Tutorials that teach you to check df.dtypes == "object" to find text columns will quietly find nothing. When a 2023 blog post and pandas 3 disagree, pandas 3 wins.

Weeks 6 and 7: SQL and DuckDB, the Half of the Job Python Does Not Cover

Our Pune claims executive now has a working notebook. She still cannot pass a first round interview, because the screening exercise is SQL, against a database she has never seen.

DuckDB is the fastest way to practise without installing a server. It runs SQL straight against CSV and Parquet files on your laptop, and the 1.4 LTS line has been maintained through 2026, with 1.4.5 released on 17 June 2026 per the DuckDB announcements. Same file as before, same answer, different language:

import duckdb

duckdb.sql("""
    SELECT city,
           sum(amount)          AS revenue,
           round(avg(amount),2) AS avg_order,
           count(*)             AS orders
    FROM 'orders.csv'
    GROUP BY city
    ORDER BY revenue DESC
""").show()

Write every week 6 and 7 exercise twice, in pandas and in SQL, then compare the outputs row by row. When they disagree you have found a real gap in joins or null handling, which is exactly what interviewers probe. The 20 SQL interview questions with model answers are good exercises once the basics hold, and the SQL optimization guide is for when your queries work but crawl.

If you enjoy this half more than the pandas half, that tells you something. People who like making data move reliably end up in data engineering, where a live Microsoft Fabric program covering DP-700 is the next step rather than a BI tool.

Weeks 8 to 10: Build One Python Data Analysis Project An Interviewer Can Poke At

Portfolio advice says build five projects. Build one, and make it survive questioning. Pick a dataset with genuine mess in it: a government release, a Kaggle CSV with three spellings of every city, or anonymised numbers from your own job. Then structure the repo so a hiring manager understands it in two minutes.

That last file is the one almost nobody writes and the one that most impresses an experienced interviewer, because it shows you audited your own work. Our Pune analyst wrote in hers that she had averaged an already-averaged column and inflated a city figure threefold. Saying that out loud reads as competence, not weakness.

6 Mistakes That Stall People Learning Python for Data Analysis

Errors you will meet in week 3, and what each one means

How to Learn Python for Data Analysis Without Losing Six Months

Free material is abundant and mostly good. The official pandas user guide, the Python documentation and Kaggle datasets will take you the whole way if you are disciplined. Google Colab removes the install problem, with the caveat that a free session runs at most 12 hours and the hardware is not guaranteed, which the Colab FAQ states plainly.

What free material will not give you is a deadline and someone to tell you your cleaning logic is wrong. That is the real argument for a live cohort, and honestly it is the only one. If you have finished things alone before, do this alone. If your last three online courses died in week two, buy the deadline. The live Data Analyst course runs Saturday and Sunday, 8:00 to 11:00 PM IST over 10 weeks with the same Python, SQL, Excel and Power BI sequence, and a free webinar or demo class costs nothing if you want to watch a session first.

Where you go next depends on which part you liked. If cleaning and pipelines pulled you in, look at data engineering and DP-700. If the modelling questions did, the MLOps engineer track covers deploying and monitoring models rather than just training them. If you caught yourself asking whether an LLM could do the cleaning, the AI engineer course covers RAG and agents, and the Copilot and agent administration course is the governance side of it. The certifications overview shows how the paths connect.

Also read: What Is Kubernetes if your analysis work starts running on someone else's cluster.

Related guides

• PL-300 Exam Prep 2026: A 6-Week Study Plan the certification most Indian analyst roles ask for once your Python basics are in place.
• Data Analyst Jobs in Pune 2026 what the market pays and who is hiring, if Pune is where you are applying.
• Power BI vs Tableau in 2026 pick the one BI tool to pair with Python before you spend a week on the wrong one.
• Data Analyst to Data Engineer in 2026 the move to make if the SQL and pipeline weeks were your favourite part.
• How to Use AI in Finance Work in 2026 prompt patterns for the reporting tasks you are automating alongside Python.

Frequently asked questions

About this guide. 360 Digital Transformation is an Authorized Training Partner of Anthropic and Microsoft. Other certification bodies, vendors and employers named here are not affiliated with us. Tools and versions change quickly; commands and figures cited were checked on 18 September 2026.