머신러닝
국제바칼로레이트 · IB Diploma · Computer Science · HL 주제 4 3:54 영어 내레이션 · 영어 + 중국어 자막 burned-in
장
전사본
A model can score well by learning a shortcut in the training data.
模型可能通过在学习数据中习得捷径而获得高分。
Evaluation must test the intended task on appropriate unseen examples.
评估必须在合适的未见示例上测试预期任务。
A relational database uses tables, keys and relationships to organize data.
关系型数据库使用表、键和关系来组织数据。
Machine learning estimates patterns from data.
机器学习从数据中估计模式。
Training and evaluation data serve different roles.
训练数据和评估数据扮演不同的角色。
Primary key is an attribute set uniquely identifying a record.
主键：唯一标识记录的属性集合。
Data leakage is the use of information unavailable at the intended prediction time.
数据泄漏：在预期预测时间不可用的信息的使用。
A primary key uniquely identifies a record, and a foreign key relates records.
主键唯一标识一条记录，外键关联不同记录。
In prediction, leakage can expose information unavailable at the real decision time.
在预测中，数据泄露可能暴露实际决策时不可用的信息。
Accuracy alone may hide an unbalanced target distribution.
仅凭准确率可能会掩盖目标分布不平衡的问题。
In the lab: Use fictional booking records to identify entities, attributes and relationships.
在实验室中，使用虚构的预订记录来识别实体、属性和关系。
For a learning exercise, use a public non-sensitive dataset, separate training and test data, and describe who may be affected by errors.
作为学习练习，请使用公开的无敏感数据的数据集，分离训练数据和测试数据，并描述谁可能受到错误的影响。
The current 2027 CS objective scope awaits the guide.
当前的 2027 年计算机科学目标范围有待指南确认。
Known: a classifier makes 90 correct predictions among 100 cases.
已知：分类器在100个案例中做出了90次正确预测。
Accuracy equals correct divided by total, times 100, which is 90 percent.
准确率等于正确数除以总数，再乘以100，即90%。
If 90 cases belong to one class, always predicting that class gives the same accuracy and can still fail every minority-class case.
如果90个案例属于同一类别，始终预测该类别会得到相同的准确率，但仍可能完全无法识别少数类案例。
Your turn: A model makes 72 correct predictions out of 80.
轮到你了。 一个模型在 80 次预测中做出了 72 次正确预测。
Find accuracy percentage.
求准确率百分比。
High accuracy is not evidence of fairness or causal understanding.
高准确率并非公平性或因果理解的证据。
A database primary key is not simply whichever field looks important.
数据库主键并非简单地由看似重要的字段决定。
Not true: A classifier with 90% accuracy must treat every class equally well.
不正确。 一个准确率为90%的分类器并不一定对每个类别都表现良好。
If 90 of 100 cases are class A, always answering A scores 90% while failing every other class.
如果100个案例中有90个属于A类，总是回答A就能得到90%的准确率，但其他所有类别都会失败。
Accuracy must be read alongside per-class results and the class balance.
解读准确率时，必须结合各类别的具体结果以及类别平衡情况。
A relational database uses tables, keys, and relationships to organize data.
关系型数据库使用表、键和关系来组织数据。
A primary key uniquely identifies a record.
主键唯一标识一条记录。
A foreign key relates records.
外键关联记录。
Watch for: high accuracy is not evidence of fairness or causal understanding.
请注意：高准确率并非公平性或因果理解的证据。
Ninety of the hundred cases belong to one class.
一百个案例中有九十个属于同一类。
A model that always answers that class is 90 percent accurate, but it gets every one of the ten minority cases wrong, so check per-class results.
总是预测这一类的模型准确率为90%，却把十个少数类案例全部判错，所以要查看每一类的结果。
A fictional dataset has 95 negatives and 5 positives.
一个虚构的数据集包含 95 个负样本和 5 个正样本。
Compare an always-negative model with one finding 4 positives but making 10 false positives. Give accuracy and positive recall.
请比较一个始终预测为负模型的准确率与召回率，以及另一个找到 4 个正样本但产生 10 个假阳性结果的模型。
The baseline accuracy is 95 percent, with recall 0 percent.
基线准确率为 95%，召回率为 0%。
The second model has 4 true positives, 1 false negative, 10 false positives and 85 true negatives: accuracy 89 percent, recall 80 percent.
第二个模型有 4 个真阳性、1 个假阴性、10 个假阳性和 85 个真阴性；其准确率为 89%，召回率为 80%。
Accuracy alone hides minority performance; evaluation design and error costs matter.
仅凭准确率会掩盖少数类表现；评估设计与错误成本至关重要。
Accuracy divides correct predictions by all cases; positive recall divides true positives by actual positives.
准确率等于正确预测数除以总样本数。 正类召回率等于真正例数除以实际正例数。
With no positive cases, recall is undefined rather than zero.
若不存在正例，则召回率为未定义而非零。
The dataset is fictional, with no trained model or personal data.
该数据集为虚构数据，不包含已训练模型或个人数据。
Retrieve the technical meanings, then vary one input in the original transfer and justify the model condition.
请回顾相关技术术语，然后在原始迁移中改变一个输入变量，并为模型条件提供依据。