Visualisation (t-SNE, UMAP)

Visualisation (t-SNE, UMAP)#

Welcome to the t-SNE & UMAP Museum of Modern Data 🖼️ This is where we take your 100-dimensional spreadsheet monster and turn it into a gorgeous 2D scatter plot that screams,

“Look! Patterns!”

🎯 Why We Need These Tools#

High-dimensional data is like a massive group photo where everyone’s talking at once. You can’t tell who’s standing next to whom — or why that one intern is in 17 dimensions.

t-SNE and UMAP help by:

  • Reducing high dimensions → 2D or 3D

  • Preserving local relationships (points that were close remain close)

  • Letting you see patterns, clusters, and chaos

Basically: They turn spreadsheets into stories. 📊➡️🎨

🧠 Meet t-SNE – “The Drama Queen of Visualisation”#

t-SNE (t-distributed Stochastic Neighbor Embedding) loves local gossip. It tries to make similar points stay close in 2D, while pushing dissimilar points far apart.

It’s nonlinear, emotional, and sometimes over-the-top — but it makes your clusters pop like fireworks. 🎆

How It Works (in drama terms):#

  1. Compute how similar each point is to every other in high dimensions (“Who are your BFFs?”).

  2. Create a 2D world where those friendships are preserved.

  3. Minimize a KL divergence loss (a fancy way of saying “less awkward rearranging”).

🪄 Example: t-SNE in Python#

from sklearn.manifold import TSNE
import matplotlib.pyplot as plt

tsne = TSNE(n_components=2, random_state=42, perplexity=30)
X_tsne = tsne.fit_transform(X)

plt.scatter(X_tsne[:,0], X_tsne[:,1], alpha=0.7, cmap='Spectral')
plt.title("t-SNE Visualization: Clusters Unleashed 💥")
plt.show()

Caution: t-SNE can look different every time — it’s like that one friend who never takes the same selfie twice. 🤳

🧩 UMAP – The Pragmatic Cousin#

If t-SNE is a passionate artist, UMAP (Uniform Manifold Approximation and Projection) is the calm data engineer. It’s faster, more scalable, and works great with large datasets.

UMAP focuses on preserving both local and global structure, so it doesn’t just gossip — it also remembers the big picture. 🧘‍♂️

⚙️ Example: UMAP in Python#

import umap
import matplotlib.pyplot as plt

reducer = umap.UMAP(n_neighbors=15, min_dist=0.1, random_state=42)
X_umap = reducer.fit_transform(X)

plt.scatter(X_umap[:,0], X_umap[:,1], alpha=0.7, cmap='plasma')
plt.title("UMAP Visualization: Order in the Chaos 🌀")
plt.show()

💡 Pro Tip: If your dataset is huge — start with UMAP. It’s fast, reproducible, and still makes your data look fabulous.

🧭 Business Applications#

Domain

Why It’s Useful

Customer Segmentation

Visualize how your clusters actually separate

HR Analytics

See employee behavior patterns in 2D

Marketing

Identify new customer groups or “weird outliers”

Product

Map user journeys or feature similarity

t-SNE/UMAP aren’t just “pretty pictures” — they help explain patterns to business folks without math anxiety. 📈💬

⚠️ Common Pitfalls#

🚩 Don’t overinterpret shapes: t-SNE is for exploration, not definitive answers.

🚩 Parameter tuning matters: perplexity, n_neighbors, and min_dist can change the visual dramatically.

🚩 Use after scaling: Always standardize your data first — otherwise, distance-based algorithms freak out.

🚩 t-SNE ≠ clustering: It just visualizes existing structure, not creates it.

🧪 Try It Yourself#

  1. Take the PCA, K-Means, or GMM dataset from earlier.

  2. Apply both t-SNE and UMAP.

  3. Compare how clusters appear in each visualization.

  4. Ask: “If I showed this to my manager, would they get the story?” (If yes, congratulations — you just made data art.) 🖌️

🧍‍♀️ Quick Recap#

Algorithm

Speed

Preserves

Use For

t-SNE

Slow

Local structure

Small, complex datasets

UMAP

Fast

Local + global

Large-scale visualization

So in short:

t-SNE paints the masterpiece, UMAP builds the gallery. 🎨🏛️

Next up: Unsupervised Lab – Customer Segmentation, where you’ll actually use PCA, K-Means, and GMM to turn chaos into marketing strategy 💼✨

# Your code here
Contents