Google

A learning system is usually optimized to fit the training examples it sees during learning, so it typically makes fewer mistakes on the training set than on the development set. As a result, the dev-error curve is usually above the training-error curve.

Training Set Error Is Usually Lower Than Dev Set Error

How do training error and dev error typically compare for a fitted model?

Validation error is usually higher than training error.

A model often makes fewer errors on the _____ set.

Match each plot element with its usual meaning.

Put the usual training-versus-validation error reasoning in order.

Question:
In a short explanation, relate a model's typical performance on the data it was fit on to its performance on held-out data, and use that to describe the positions of the two error curves.

Sample answer:
A model is usually optimized to perform best on the training data. That usually makes the training error lower than the validation error. Since lower error is drawn lower on the graph, the validation-error curve usually sits above the training-error curve.

Key points:
- The model usually performs better on the training data.
- Better performance means lower error.
- Lower error places the training curve below the validation curve.

Rubric:
A strong response should mention the typical performance gap, connect that gap to lower training error, and explain why the validation curve appears higher.

Why the validation-error curve is usually above the training-error curve

Case context:
A learning curve chart shows training error in teal and validation error in orange. Across the full plot, the orange line remains higher than the teal line.

Question:
Is this the expected pattern for training and validation error, and what does it say about how the model performs on the two data sets?

Sample answer:
Yes. This is the usual pattern. When validation error is above training error, it means the model fits the training set more closely than the validation set. In other words, performance is better on the training data than on the held-out data.

Key points:
- Orange is the validation-error curve.
- Teal is the training-error curve.
- The validation curve is higher throughout the chart.
- The model performs better on the training set than on the validation set.

Rubric:
The response should recognize the pattern as normal and correctly explain that a higher validation-error curve means the model is doing better on the training data.

Read a chart where the validation-error line sits above the training-error line.

Question:
Answer in one to three sentences using the usual relationship between training and development errors.

Sample answer:
It means the development error is larger than the training error. The model is doing better on the training examples than on the development examples, which suggests weaker generalization.

Key points:
- Development error is higher than training error.
- The model fits the training set more closely than the development set.

Rubric:
The answer should mention both the higher development error and the model's better performance on the training set.

Learn Before

Related