Liverpool Videos

The Albert
844
1 Like
845
846

847

https://x.com/TheRedmenTV/status/1926342515256336891

https://x.com/TheRedmenTV/status/1926336842724495744

2 Likes
848

https://x.com/luchogarcia14/status/1926448572750258372

3 Likes
849

https://x.com/BBCMOTD/status/1926542074934931515

3 Likes
850

851
1 Like
852
1 Like
853

https://x.com/bbcmerseysport/status/1927050405432734082

854

https://x.com/taggysbar/status/1927296187918438580

855

https://x.com/NBCSportsSoccer/status/1927734706436554927

856
857
858

Data Guru Ian Graham on the analytics behind Liverpool’s title win & Salah and VVD’s new contracts.

Liverpool’s original data analytics wizard, Ian Graham, joins Rog for an in-depth discussion on how Liverpool used data analytics to build a Premier League and Champions League winning team. Ian also discusses the boom in analytics in football and talks about the future of data analytics in the sport and why some teams are still hesitant to embrace it.

1 Like
859

I was reading some background on vector databases last week. On the face of it, it doesn’t look radically different to things I was doing 40 years ago (at least mathematically) but I was curious how these sorts of things can be applied to football.

For example, is it possible to spot a player that is underperforming because they are being played out of position but might fit in elsewhere?

I can recall several occasions when Klopp said that he had “tweaked” his tactics to suit one of his players.

860

You’re absolutely right, mathematically, the core concepts behind vector databases and embeddings (like measuring similarity in high-dimensional space) are not new. What is new is the scale, speed, and flexibility with which we can now compute and store this information, especially thanks to advances in machine learning, vector search, and high-performance storage systems.

Now, to your core question: Can these ideas be applied to football to spot misused players or optimize their positioning?

Using Vectors in Football Analytics

At the heart of modern football analytics is player representation—turning a player’s behavior, performance, or style into a vector (i.e., a list of numbers). Once you do that, you can:
• Compare players using distance/similarity measures.
• Cluster players into roles or styles.
• Predict compatibility with systems or positions.

  1. Player Embeddings

You can embed players based on various data types:
• Positional data: heatmaps, movement patterns, average location per phase.
• Event data: passes, tackles, dribbles, pressures, etc.
• Tracking data: acceleration, spacing, off-ball runs.
• Tactical fit: how a player behaves in certain formations/phases.

These embeddings can be compared using cosine similarity or nearest neighbor search via a vector database like FAISS or Pinecone.

Practical Application: Positional Misuse

Let’s say you have a winger who’s underperforming. Here’s how embeddings could help:

Step 1: Build a Vector Profile

Collect in-game data for the player and represent their play style as a vector.

Step 2: Search for Similar Vectors

Use a vector database to find players with similar vectors but in different positions—maybe they’re all inside forwards or attacking midfielders.

Step 3: Tactical Matching

Cross-reference the player’s vector with tactical setups (e.g. Klopp’s 4-3-3 vs 4-2-3-1) to see which system styles historically matched players like him.

Step 4: Recommend Role Shift

If embedding similarities point strongly toward a different role, it gives evidence to suggest trying the player elsewhere.

Klopp’s Tactical “Tweaks”

When Klopp says he “tweaked” his system, he might be:
• Moving a player inside/outside
• Shifting pressing triggers
• Adjusting build-up shape to allow someone to receive more comfortably

Vector models could be used to simulate these tweaks:
• You run a simulation of “if this player is given a different heatmap role, what does his similarity to other successful players look like?”
• Or ask, “does his vector move closer to high-performing players in a slightly different position?”

Real-World Examples
• Trent Alexander-Arnold’s hybrid role: Early models based on heatmaps and passing vectors suggested he resembled midfield playmakers more than traditional full-backs. Klopp eventually trialed him in a hybrid role.
• Kai Havertz at Arsenal: Struggled as a false 9, improved as a left 8. Vector models might’ve suggested this based on positional comparisons.

If you had this data at scale, across thousands of players, a vector database (like Pinecone or FAISS) would:
• Store each player’s vector.
• Let you instantly query for “most similar players” to anyone.
• Filter by role, league, team, etc.

So you could ask: “Find me players in the Bundesliga whose style matches this underperforming winger in the Eredivisie,” and it could help with scouting or repurposing.

1 Like
861

Thanks for the reply. I suspect that I will be disappearing down a statistical rabbit hole on Monday.

862

FIFY

863