Using Uneven Training Sizes to Save Time on Learning Curves
When training many models for a learning curve is expensive, it is often enough to test a handful of unevenly spaced dataset sizes. For example, using 500, 1,000, 2,000, 5,000, and 8,000 examples can reveal the overall shape of performance while requiring fewer full training runs than checking every equal step.
0
1
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
Using Uneven Training Sizes to Save Time on Learning Curves
What belongs on the vertical axis of a learning curve built from different training-set sizes?
A learning curve is built by training one model on the full training set and then reading off performance at different checkpoints during that same run.
To make a learning curve, you train _____ versions of the model on training sets of different sizes.
Match each learning-curve element to its job in the experiment.
Arrange the steps for building a learning curve from a labeled training pool of 800 examples.
Which procedure correctly builds a learning curve from a fixed pool of 1,200 labeled examples?
Using the Same Validation Set for Subsampled Training Runs
Learning-curve axes
Match each training subset description to its role on a learning curve built from 1,200 total examples.
Arrange the steps for building a learning curve from multiple training sizes
How to build a dev-set error curve from different training set sizes
Plan a learning-curve experiment with 800 labeled examples.
What metric goes on the y-axis of a learning curve built from different training-set sizes?
Learn After
Why sample training set sizes at uneven intervals?
Smaller Training Sets Usually Train Faster
Reducing Cost with Nonlinear Data Sizes
Choosing Dataset Sizes for Learning Curves
Choosing Spacing for Learning-Curve Experiments
Why Uneven Training Sizes Save Time
Selecting Sample Sizes for an Efficient Learning Curve
When are uneven training-set sizes useful?
Why Use Uneven Training-Set Sizes for a Learning Curve?
You should always use evenly spaced training-set sizes, even when training extra models is expensive.