A horse recently (and inaccurately) predicted the United States would beat Germany in their World Cup match up last week.
Obviously, that didn't pan out.
It's unlikely that anyone went back to the equine seer ahead of the knockout round against Belgium on Tuesday, but there was at least one forecaster that got it right.
That would be Google.
The Internet giant is touting how the Google Cloud Platform has been crunching large datasets and statistics (many of which flood Twitter with every kick, goal and penalty in each match) to predict the outcome of each game.
Starting with numbers generated by live sports data firm Opta, Google engineers have employed an amalgam of Google Cloud Dataflow to ingest data, BigQuery to build derived features, iPython and Pandas to conduct modeling, and finally the Compute Engine to crunch the data.
This evolved into a logistic regression approach in predicting the winners versus a poisson regression, according to Mountain View.
And so far, it seems to be working as the Google Cloud Platform has boasted a perfect record in the nail-biting Round of 16.
However, Google engineers also acknowledged there weren't any "any major upsets" this last round either.
For anyone who wants to take a chance on Google Cloud Platform's method, here's a glance at the predictions for the quarterfinals this weekend:
- Brazil vs. Colombia: Brazil (71%)
- France vs. Germany: France (69%)
- The Netherlands vs. Costa Rica: Netherlands (68%)
- Argentina vs. Belgium: Argentina (81%)