Open WorkBuddy and select the configured Six-Loop collaboration team (六环成长协作团). Submit the task pack without test labels first. Record predictions, then provide the labeled JSON. If you have installed daily-compounding-sixloop, use it to capture your first question, why you chose a model, and what you need to verify next.
View the full task for copying
PBL-002 · v1.0.0
Hotel booking cancellations: how do you choose when accuracy is tied?
1 = canceled, 0 = not canceled. All records are simulated.
Predict cancellation using only lead days, which are known at booking. A cancellation timestamp is unavailable then and would leak the outcome.
The training set H01–H08 comes from the original exercise. The separate test set T01–T08 is fixed. Keep its labels hidden until predictions are recorded and never use it to select parameters.
KNN uses k=3, absolute distance in lead days, training booking ID to break distance ties, and a majority vote.
The preset baseline predicts no cancellation for every booking. The training labels are evenly split.
Training examples with labels
Booking, Lead days, Actual label
H01, 2, 0
H02, 4, 0
H03, 7, 1
H04, 10, 0
H05, 16, 1
H06, 20, 1
H07, 24, 0
H08, 28, 1
Separate test bookings with labels hidden
Booking, Lead days
T01, 3
T02, 6
T03, 8
T04, 10
T05, 12
T06, 17
T07, 22
T08, 27
Assign independent checks in WorkBuddy
Open WorkBuddy and select the configured Six-Loop collaboration team (六环成长协作团). Submit the task pack without test labels first. Record predictions, then provide the labeled JSON.
If you have installed daily-compounding-sixloop, use it to capture your first question, why you chose a model, and what you need to verify next.
1. Have the lead restate the problem, acceptance criteria, and assignments. The question coach describes the decision and the consequences of false alarms and missed cancellations.
2. Have the learning designer calculate both sets of predictions and return the predictions, neighbors, and method. The evidence reviewer independently checks the split, prediction time, and tie ordering. Wait for test labels until the predictions are recorded.
3. Once the labeled JSON arrives, have the evidence reviewer compute both confusion matrices and metrics from the saved predictions. Mark precision undefined when there are no predicted positives. Preserve the original predictions and k.
4. The lead keeps independent reports and disagreements for the teacher. Check leakage, split integrity, false alarms versus missed cancellations, the next evaluation, and the limits of simulated data.
https://www.treeye.cn/en/problem-competency#pbl-002