Juditis Mission - 11/12

Context

Following a cosmic phenomenon called the Chaos Convergence, characters from multiple comic book universes have been reunited into a single reality. The Interdimensional Hero Monitoring Agency (IHMA) has centralized data about these heroes to quickly analyze their profiles and make automated decisions.

The recorded data describes both general character traits and associated abilities and numerical statistics. Analyzing this data is important for:

• quickly extracting descriptive statistics about the dataset;
• reconstructing missing information for new characters;
• classifying abilities based on other available traits.

To solve the problem, you will have two files at your disposal:

• train.csv – contains labeled examples used for analysis and model training;
• test.csv – contains the same variables but does not include the powerstats__combat and Super Strength columns, which must be predicted in Task 3 and Task 4.

Hero Universes

The dataset includes characters from several editorial universes, for example:

• Marvel Comics
• DC Comics
• Dark Horse Comics
• Image Comics
• Shueisha
• George Lucas
• Microsoft
• J. K. Rowling

In this problem, the notion of "universe" corresponds to the value in the Publisher column.

Dataset Description

Each row in the train.csv and test.csv files corresponds to a character.

The dataset contains:

• a unique identifier: id;
• descriptive variables about the character;
• binary variables indicating the presence or absence of certain abilities;
• continuous numerical variables of type powerstats__...;
• additional columns with information about heroes.

Example columns:

Requirements

Task 1 - Number of Distinct Universes (10 points)

Using the data from the train.csv file, determine how many distinct values appear in the Publisher column.

Task 2 - Publisher with the Most Positive Heroes (10 points)

Using the data from the train.csv file, consider only characters where Alignment = good.

Determine which publisher contains the most positive heroes.

In case of a tie, choose the publisher that comes first alphabetically.

Task 3 - Prediction of powerstats__combat (40 points)

Based on the data from the train.csv file, train a model capable of predicting the values of the powerstats__combat attribute for each character in test.csv.

Task 4 - Prediction of the Super Strength Ability (40 points)

Based on the data from the train.csv file, train a model capable of predicting the values of the Super Strength attribute for each character in test.csv.

Allowed values are (predictions must be final binary labels, not probabilities):

10.021.0

Submission File Format

The submission file must be in csv format and contain the following columns:

Possible values for subtaskID are:

For Task 1 and Task 2, the answer is global and is submitted using the special value GLOBAL in the id column.

Example submission.csv file

1id,subtaskID,answer2GLOBAL,task1,13GLOBAL,task2,Cartoon Network443,task3,2.57550,task3,5.1761,task4,0.074,task4,1.0

Evaluation

The total score is 100 points.

Task 1 (10 points)

Evaluation is done by exact comparison between the submitted value and the correct answer.

Task 2 (10 points)

Evaluation is done by exact comparison between the submitted value and the correct answer.

Task 3 (40 points)

For Task 3, the MAE (Mean Absolute Error) metric is used.

Points are awarded as follows:

• if MAE ≤ 12.0, 40 points are awarded;
• if MAE ≥ 30.0, 0 points are awarded;
• for intermediate values, the score is linearly interpolated.

The formula used is:

1score_task3 = 40 × (30 - MAE) / (30 - 12), for 12 < MAE < 30

Task 4 (40 points)

For Task 4, the Accuracy metric is used.

Points are awarded as follows:

• if Accuracy ≥ 0.90, 40 points are awarded;
• if Accuracy ≤ 0.55, 0 points are awarded;
• for intermediate values, the score is linearly interpolated.

The formula used is:

1score_task4 = 40 × (Accuracy - 0.55) / (0.90 - 0.55), for 0.55 < Accuracy < 0.90