Choose and Explain Models checkpoint

Supplied for this practice

These objects are prepared afresh before your editor runs.

group_sizes: Supplied group_sizes; its definition below specifies the values and role.
group_sizes = pd.Series([300, 150, 50], index=['group 0', 'group 1', 'group 2'])

Outputs you create

rmse_difference: Requested result described in the task and deliverables.
minority_missed: Requested result described in the task and deliverables.
group_proportions: Requested result described in the task and deliverables.
unsampled: Requested result described in the task and deliverables.
omitted_from_display: Requested result described in the task and deliverables.

Exercise evidence · ML-M-K1-1 · all 2 rows
`wine_rmse`	`minority_actual`	`minority_caught`	`profile_rows`	`population_rows`	`retained_variance`	`shown_variance`
0.8	10	2	500	569	0.9	0.65
0.81	10	2	500	569	0.9	0.65

Your task

Check the five cases below.
Write a short report for each case.
Wine: candidate A has CV RMSE 0.8.
Candidate B has CV RMSE 0.81.
Fold variation is 0.1.
Penguins: classifier accuracy is .95.
Recall for the minority class is .2.
K-Means: k=3 has silhouette .51. k=4 has silhouette .52.
Their group sizes are very different.
Ward: the profiles use a 500-row sample.
The report claims they describe all 569 rows.
PCA: seven axes keep 90% of the variance.
The picture shows only two axes.
For each case, name the task.
Explain what the evidence supports.
State one limitation.

rmse_differenceminority_missedgroup_proportionsunsampledomitted_from_display

Hints

Think: The wine score gap is small relative to the stated fold variation of .1. Minority errors, group sizes, sample scope and display retention answer different questions; they cannot form one accuracy leaderboard.

Tools: Choose and Explain Models checkpoint: use the supplied wine_rmse, minority_actual, minority_caught, profile_rows, population_rows, retained_variance, shown_variance columns and the pandas or model operations introduced in this lesson.

Approach: Compute rmse_difference, minority_missed, group_proportions, unsampled, omitted_from_display from the supplied evidence. Preserve the requested labels and row order, then inspect the values before checking.

Explained solution

rmse_difference = df['wine_rmse'].iloc[1]-df['wine_rmse'].iloc[0]
minority_missed = df['minority_actual'].iloc[0]-df['minority_caught'].iloc[0]
group_proportions = group_sizes/group_sizes.sum()
unsampled = df['population_rows'].iloc[0]-df['profile_rows'].iloc[0]
omitted_from_display = df['retained_variance'].iloc[0]-df['shown_variance'].iloc[0]
print(rmse_difference)
print(minority_missed)
print(group_proportions)
print(unsampled)
print(omitted_from_display)

The wine score gap is small relative to the stated fold variation of .1. Minority errors, group sizes, sample scope and display retention answer different questions; they cannot form one accuracy leaderboard.