BRAINCheck · CT & AI LAB · ORIGINAL COMPANION WORKSHEET

Training से अलग testing

Keep an unseen test set to check a model.

Name: ____________________ Class: ______ Date: __________

Performance on examples already seen during training can overstate how well a system handles new examples. Reserve a test set before practising a rule. Do not silently move difficult test items into training and continue to call the old test score independent. Record both successes and failures.

A classifier got all six training cards right. Its maker claims: “It will classify every new card correctly.”

Create ten fictional object cards and reserve four without looking at their answers while designing a rule.
Design a rule using the other six.
Test it on the reserved cards and record correct/total.
State what the test result supports and what it cannot guarantee.

Quick check

Which set gives a more independent check after training?

○ The same examples memorised during training

○ A separate previously unseen test set

○ Only examples chosen after seeing their answers

○ No examples

My correction / justified conclusion

Evidence and reference actually checked

My reasoning, uncertainty and reflection

Use fictional/non-personal examples. Public worksheet; save or submit work only through your own notebook. Suggested 60-minute activity. Not an official NCERT/CBSE assessment or certificate.