Deep Learning with PyTorch

Measure of success

The measure of success will be directly determined by your business goal. For example, when trying to predict when the next machine failure will occur in windmills, we would be more interested to know how many times the model was able to predict the failures. Using simple accuracy can be the wrong metric, as most of the time the model will predict correctly when the machine will not fail, as that is the most common output. Say we get an accuracy of 98%, and the model was wrong each time in predicting the failure rate—such models may not be of any use in the real world. Choosing the correct measure of success is crucial for business problems. Often, these kinds of problems have imbalanced datasets.

For balanced classification ...