A compact EDA implementation pattern

This pattern assumes the course CSVs have been loaded and passed is a Boolean/0-1 outcome. The CSV files were not attached, so use it as a code-reading template rather than as a claim about an executed dataset here.

joined = enrolments.merge(
    students[['student_id', 'study_mode']],
    on='student_id', how='left', validate='many_to_one'
)
assert len(joined) == len(enrolments)

def rates(group_columns):
    out = joined.groupby(group_columns)['passed'].agg(
        registrations='size', known='count', passes='sum'
    )
    out['missing'] = out['registrations'] - out['known']
    denom = out['known'].where(out['known'].gt(0))
    out['known_pass_rate'] = out['passes'] / denom
    return out

pooled = rates(['study_mode'])
within_course = rates(['course_code', 'study_mode'])

An all-missing group has no known-outcome rate; do not call its zero observed successes a 0% pass rate. Inspect matching and missing group labels as well. If evaluating predictions for new students, keep each student's registrations in one split.

Lab 5's lecture handoff asks for an explicit question and population, checked join keys and missing n, pooled versus within-course rates, and an observation, explanation and limitation. No separate Lab 5 notebook was attached.