Ricardo
Vieira

A public analytics platform for the Liga Portugal, covering seven seasons of event data. I do all of it: the ingestion, the models on top of it, and the match, team and player reports that make the output usable.

The ingestion pipeline holds 2.8 million events across 1,872 matches, reconciled from two independent providers by fuzzy entity matching. The matchday sync runs unattended, with backfills chained behind it and recovery for events that arrive incomplete.

• An expected pass model that gives every pass a completion probability from its origin, destination, pass type qualifiers and game state. Validated on held-out matches, with calibration checked per pitch zone and separately for open play and set pieces rather than on one aggregate figure.
• Role clustering that groups players by how they actually play, from touch locations and per 90 output rather than the position they are listed at. It surfaces players whose listed position hides what they contribute, the kind of mismatch a position filter misses in recruitment screening.
• An expected points table from a double-Poisson model over match xG, separating results from underlying performance in the standings.
• The analyst-facing output: match reports, player and team profiles, head to head, percentile charts against position and role peer groups, pass sonars, pass zone and box-entry maps, shot maps and similarity search.

• pitchwatch.rjvieira.com

A metric measuring how versatile a player is, from which of seven action types they complete successfully in each of six functional pitch zones, normalised per minute played and bounded between 0 and 1. Applied to Opta event data from the big five leagues and the Primeira Liga, 2024/25, and published as an open source Python package.

Each action type is weighted by the inverse of its 95th percentile rather than by z score, because the underlying actions are not normally distributed and rare ones like shots on target would otherwise count for almost nothing. We tested it against a Shannon entropy version of the same idea, to check whether versatility is better captured by the diversity of a player's actions or by how evenly they are spread across the pitch. Zones, action set and weights are all parameters, so it runs against other datasets.

• tvi-footballindex on PyPI

pip install tvi-footballindex

import pandas as pd
from tvi_footballindex.parsing import f24_parser
from tvi_footballindex.tvi import calculator

events = f24_parser.parsef24_folder("data/F24")
play_time = f24_parser.calculate_player_playtime(events, min_playtime=30)

actions = pd.concat([
    f24_parser.get_interceptions(events),
    f24_parser.get_tackles(events),
    f24_parser.get_aerials(events),
    f24_parser.get_progressive_passes(events),
    f24_parser.get_dribbles(events),
    f24_parser.get_key_passes(events),
    f24_parser.get_deep_completions(events),
    f24_parser.get_shots_on_target(events),
])

tvi_df, quantiles = calculator.calculate_tvi(actions, play_time)
# TVI, TVI_entropy and shannon_entropy, per player per game

The basketball one. A web app for understanding what happened in an NBA game, past what the traditional box score puts on the page.

• boxscorewatch.rjvieira.com
• Source

My first go at football match analysis, before the postgraduate degree. Explore match stats and export the visualisations as images. Pitchwatch is where all of this ended up.

• soccerstats.rjvieira.com
• Source