Essay

Analyzing the root cause when training error is 10% and dev error is 12%

Question: In a scenario where an algorithm achieves a 10% training error, an 11% training-dev error, and a 12% dev error, analyze what this specific combination of error metrics indicates about the model's performance on the training set. Why does this pattern lead to this conclusion?

Sample answer: This combination of error metrics indicates that the algorithm has high avoidable bias. Because the training error is relatively high (10%) and the subsequent gaps to the training-dev error (1%) and dev error (1%) are small, the primary issue is that the algorithm is doing poorly on the training set itself, rather than struggling primarily with variance or overfitting.

Key points:

  • Identifies high avoidable bias.
  • States the algorithm is doing poorly on the training set.
  • Notes the small gaps between error metrics indicate variance is a secondary concern.

Rubric: Answers should identify high avoidable bias and explain that the core problem is poor performance on the training set, noting that the gaps between the training, training-dev, and dev errors are small compared to the initial training error.

0

1

Updated 2026-06-12

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI