What if I do not have a dataset?

A few workable public ones are suggested, but collecting forty rows yourself is genuinely part of the sprint. Data you gathered is data whose quirks you understand, and understanding the quirks is most of what makes the write-up worth doing.

Is two weeks enough to learn clustering?

Enough to run it competently and to be appropriately sceptical of the results, which is the useful bar. Choosing between clustering algorithms for a particular data shape is a deeper topic and sits in the Classical ML phase course.

Does this need a fast computer?

No. Everything here runs in your browser on datasets of a few hundred rows, which is the right scale for learning — you can see every point, which you cannot do at a million rows.

Build sprints₹299

Find the groups in your own data

Take this if you have a spreadsheet of something you care about and suspect there are patterns in it.

What you’ll be able to do

You can cluster a dataset, choose the number of clusters with a defensible reason, reduce it to two dimensions to look at it, and recognise when the clusters you found are an artifact rather than a finding.

What it assumes

Basic Python and a spreadsheet you can export as CSV. No prior machine learning.

Price: ₹299
Length: 2 weeks
Sessions: 6
Effort: ~4 h/week · 8 h total
For: Class 9–10 · Class 11–12 · College · Working professional

Adds it to My courses on your dashboard so you can find it again. Payment is not open yet — nothing is charged.

What’s taught, in order

01
Grouping without labels
What changes when nobody tells you the right answer, and why that makes evaluation genuinely hard rather than merely different. This framing prevents the standard mistake of trusting whatever comes out.
02
k-means, driven by hand
Stepping the algorithm one iteration at a time in the visualiser, watching centroids move, and re-running it from a different start to find that the answer changes. Instability is a property, not a bug.
03
Choosing k without fooling yourself
The elbow method, its silhouette alternative, and the honest conclusion that both are heuristics. Learning to say the data does not clearly separate is part of the skill.
04
Looking at it in two dimensions
PCA as the lens that lets you see a many-column dataset on a page, plus what the projection is throwing away and why a clean-looking plot can still mislead.

What you build

Collect a dataset of at least forty rows and four numeric columns about something real to you — your own spending, your class's test scores, weather, cricket statistics. Cluster it, project it to two dimensions, and write up what each cluster represents in plain language plus one cluster you do not believe and why.

Check it yourself against this

The data is genuinely yours and has at least forty rows and four numeric columns.
Your choice of k is justified with a plot or a metric, not just picked.
Each cluster is described in the vocabulary of the subject, not as cluster 0 and cluster 1.
You named one cluster you do not trust, with a specific reason.
You can say what the PCA projection is hiding.

Then answer this

k-means will happily split perfectly uniform random data into three neat clusters — so what makes a clustering real?

Two or three sentences, in your own words. If you cannot, the course is not finished — go back to the module it came from.

What this course does not do

k-means only, so the density-based and hierarchical methods that handle awkwardly shaped clusters are out of scope, as is any clustering of text or images. It also cannot tell you whether your groups mean anything in the world; that judgement stays yours.

Where this leads

The k-means visualiser is free and standalone, and re-running it from different starting positions on the same points is the quickest way to see why clustering results have to be treated with suspicion.

Step through k-means Try a real lesson free

FAQ

What if I do not have a dataset?: A few workable public ones are suggested, but collecting forty rows yourself is genuinely part of the sprint. Data you gathered is data whose quirks you understand, and understanding the quirks is most of what makes the write-up worth doing.
Is two weeks enough to learn clustering?: Enough to run it competently and to be appropriately sceptical of the results, which is the useful bar. Choosing between clustering algorithms for a particular data shape is a deeper topic and sits in the Classical ML phase course.
Does this need a fast computer?: No. Everything here runs in your browser on datasets of a few hundred rows, which is the right scale for learning — you can see every point, which you cannot do at a million rows.

Other courses

This is a standalone course. It does not add weeks to the 78-week curriculum or change your roadmap. If you want the whole path instead, the full programme is ₹399/month · ₹2,000/year — see what it covers.