Script 3 feedback notes

from #43 (closed)

• first exercise section, pt.2 => it isn't but should be lower for low k-folds (cf. C=1)
• last exercise section, pt 2. => should be not for mnist data, but for the previously used dataset