• Ball-by-ball data from 1600 innings of the past T-20 matches.

• Sample input file which the code has to accept and process.

• For each innings of each match, the output of the code must be a single number, which is the score at the end of the 6th over.

• Remove extra columns from the data and keep only the following:

    1. venue
    2. innings
    3. ball
    4. batting_team
    5. bowling_team
    6. striker
    7. non_striker
    8. bowler
    9. runs_off_bat
    10. extras

• Calculate total runs as sum of runs_off_bat and extras.
• Keep data whose ball < 6, discard the rest of the rows. This is because our aim is to find the total runs only upto 6 overs, which makes other balls irrelevant to the problem statement.
• Keep data where innings <= 2, discard the rest of the rows
• Create a new column called batsmen as a list of values from columns striker and non_striker. This is because the test input only provides a single column with all batsmen in the 6 overs.
• Group the rows on match_id using the following aggregation functions for each of the columns:

  1. venue - first
  2. innings - first
  3. batting_team - first
  4. bowling_team - first
  5. total_runs - sum
  6. batsmen - list
  7. bowler - list

• Create a list of unique batsmen from the column batsmen as batsmen_all
• Similarly create bowler_all
• Create two dictionaries batsmen_dict and bowler_dict with a digit to represent each of the unique values from batsmen_all and bowler_all, respectively. This is done for encoding the batsmen and bowlers.
• Create and store pkl files of batsmen_dict and bowler_dict to use for the test set later
• Replace the batsmen column values with the corresponding encodings from batsmen_dict, similarly for bowler column values, using bowler_dict
• Split the batsmen ecoded list values into separate columns as batsmen_0, batsmen_1, etc. Similarly split for bowler into bowlers_0, bowlers_1, etc.
• Drop the batsmen and bowler columns as the values have already been inserted as new columns.
• Create two objects of LabelEncoder() as venue_encoder and team_encoder
• Fit the venue, batting_team, bowling team using the encoder objects.
• Encode the venue, batting_team, bowling team columns using the fit encoders
• Store the fit encoders for later use during testing
• Split the data into training and testing, with testing being column total_runs and training being rest of the columns apart from total_runs
• Use sklearn’s train_test_split to split the data into training and testing sets: x_train, x_test, y_train, y_test
• Create an object of StandardScaler() and fit on the x_train data, use it to transform the x_test data
• Create a object of MinMaxScaler() and fit the y_train data, use it to transform the y_test data
• Store both the x data scaler as well as y data scaler for later use.

• Create a keras Sequential model object
• Model is created as the layers shown below
• Fit the model on x_test and y_test with batch_size=10 and epochs = 100
• Store the model for training purpose

• Use the trained model on x_test and store the results

Figure 1: Model Summary

Figure 2: Model Training

• Perform inverse transform on y_test and y_pred, to obtain the original values for comparison using the same MinMaxScaler object we created before.
• Plot the data to view the results

Figure 3: Validation

• Read the test input
• Read the batsmen and bowler encoding dictionary as batsmen_dict and bowlers_dict
• Replace the batsmen column with the corresponding encoding values from batsmen_dict, for new values suppply new encoding values. Perform similarly for bowlers using bowlers_dict
• Split the encoded list values in bowlers and batsmen columns into new lists as bowlers_0, bowlers_1 etc and similarly for batsmen_0, batsmen_1 etc.
• Drop the batsmen and bowlers columns
• Read the previously stored venue and team encoders. Use them to encode the venue, batting_team and bowling_team columns
• Ensure the order of columns, and their names are as below and fill empty values with zero

Figure 4: Validation

1. venue
2. innings
3. batting_team
4. bowling_team
5. batsmen_0
6. batsmen_1
7. batsmen_2
8. batsmen_3
9.  batsmen_4
10. batsmen_5
11. batsmen_6
12. batsmen_7
13. batsmen_8
14. batsmen_9
15. batsmen_10
16. batsmen_11
17. batsmen_12
18. bowler_0
19. bowler_1
20. bowler_2
21. bowler_3
22. bowler_4
23. bowler_5
24. bowler_6
25. bowler_7
26. bowler_8
27. bowler_9

• Read the stored Scaler object and use it to transform the data. Store it as x_test.
• Read the stored Keras Sequential Model and use it to predict the scores with x_test
• Load the stored Target Scaler and inverse transform the prediction of the model
• Return the prediction to the main function as result