Supervised vs Unsupervised: The ML "Market Map"
Learn the difference between AI that predicts a known answer and AI that just finds structure, and what the market map does.
Category: Machine Learning
In plain English
Machine learning is a way of letting a computer find patterns in data on its own. There are two big styles. Supervised learning has a 'right answer' to copy: you show it many examples that are already labelled (this email is spam, that one is not) and it learns to label new ones. Unsupervised learning has no answer key at all: you just hand it the data and it groups things that look alike or points out the odd ones. SurgeFlow's ML 'market map' is the second kind: it sorts stocks into look-alike groups and flags unusual ones, without ever being told what 'good' or 'bad' means.
What it measures
Unsupervised learning measures similarity, not future price. The market map looks at each stock's recent numbers (things like how much it moved, how heavily it traded, its size) and asks 'which stocks look statistically alike today?' Stocks that look alike land in the same group, or 'cluster'. A stock whose numbers look unlike everything else is flagged as an 'anomaly' (an outlier). - Supervised example: from 1,000 labelled emails it learns to tag a new email 'spam' or 'not spam' — there was a correct answer to match. - Unsupervised example: given 500 stocks with no labels, it might form 5 clusters — say 200 'big and calm', 150 'small and jumpy', and so on — purely from how their numbers sit together. - Anomaly example: if 499 stocks traded near their normal volume but 1 traded 20x its usual amount, that 1 gets flagged as unusual.
How SurgeFlow uses it
This underpins the ML 'market map' tab. Once a day, for all 4 markets (US, China, Japan, Hong Kong), SurgeFlow takes a batch of features for each stock, puts the numbers on a common scale, and runs clustering to form the groups plus an outlier check to flag unusual names. The result is a snapshot of how stocks were arranged on that day. It is descriptive: it summarises today's structure of the market, and is rebuilt fresh each day.
How to read it on the site
Open the ML tab (/ml). You will see stocks placed into clusters — each cluster is just a bucket of stocks that looked similar on that day — and some names marked as anomalies. Read a cluster as 'these behaved alike recently', and an anomaly flag as 'this one stood out, worth a closer look', not as a verdict. The groups can shift from one day to the next because the map is rebuilt daily from fresh numbers. - A cluster = a group of look-alike stocks for that day; the cluster has no built-in 'better' or 'worse' meaning. - An anomaly flag = this stock's recent numbers were unusual versus the crowd — a prompt to investigate, not a signal. - Group membership can change day to day; that is normal, not a glitch.
What it cannot tell you
The market map describes structure; it does not predict prices and never says buy or sell. Being in a 'calm' cluster does not mean a stock will stay calm, and an anomaly flag does not mean a stock will rise or fall — outliers can be unusual for boring or one-off reasons. Because there is no answer key, the groups are not 'right' or 'wrong', only one reasonable way to arrange today's data; a different setup could draw the lines differently. Treat it as a map of where things sat today, then do your own homework — SurgeFlow is research and education, not investment advice. - No forecasts: similarity today says nothing certain about tomorrow's price. - Not a signal: cluster or anomaly is not a buy/sell instruction. - No ground truth: with no labels there is no single correct grouping — it is descriptive only.