liveBook · Manning

Mean absolute error

As explained in Build AI Drug Discovery Pipelines

Mean absolute error is the average of the absolute differences between observed target values and model predictions.

Mean absolute error (MAE) is the average absolute difference between observed values and model predictions. For each prediction, take the absolute value of its error, then average those values. Lower MAE indicates a better fit. section 4.1.4

MAE can be reported alongside mean squared error (MSE). Unlike MAE, MSE squares each error, placing greater emphasis on larger errors. Interpret either score in the context of the target values and the intended use of the model. section 4.1.4 section 8.1.2

Evaluating kinase activity predictions

A PyTorch model predicting pIC50 values reports a test MAE of 0.7274 and an MSE of 0.9249. The MAE means the predictions differ from their observed values by 0.7274 pIC50 units on average. These tutorial results alone are not enough to justify screening decisions. section 8.1.2

Test Loss: 0.9235
Mean Squared Error: 0.9249
Mean Absolute Error: 0.7274
 

Inspect the reported test loss, MSE, and MAE together. section 8.1.2

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines MAE as an aggregate of absolute residuals and reports a test result; defers evaluation-function implementation details.
Examples
Reports a kinase model’s test MAE of 0.7274 alongside its mean squared error.
What you can do after
Evaluate/understand model error using MAE alongside other metrics and the target-value range.
How it connects
Connects MAE to residuals, mean squared error, linear regression, and test-set evaluation.

Also explained in

Compare how 15 books explain mean absolute error →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.