To use previous IPL match data to simulate a match given:

• Teams
• Batting order
• Bowling order
• Toss decision

The idea is to generate probabilities of scoring 0, 1, 2, 3, 4, 5, or 6 runs or of a wicket for every batsman-bowler pair and simulate the score of the match ball-by-ball.

For handling unseen batsman-bowler pairs, we use two approaches:

Implemented using Hadoop MapReduce, we cluster together similar batsmen and bowlers using certain parameters and then generate probabilities for each cluster-cluster pair.

This method uses Spark's MLLib to perform Collaborative Filtering. It uses recommendation systems to generate values to fill empty batsman-bowler pair data.

• The first files to be generated are the data files using the codes in /data/
• All the code is available in /src/ . The way to execute is, execute ./run.sh from both, the batsmen and bowler directories. This runs the k-means code using MapReduce.
• Match simulation code is in /src/probcalc/.
• First change Team1bats.csv, Team1bowl.csv, Team2bats.csv, Team2bowl.csv to Team 1 batting order, Team 1 bowling order, Team 2 batting order and Team 2 bowling order respectively in both, /CollaborativeFiltering/ and /KMeansClustering/.
• Run run.sh in /CollaborativeFiltering/ and simMatch.py in /KMeansClustering/ to simulate matches using both strategies.
• This will give you the ball-by-ball simulated output for every match.