7.14 Takeaways

This chapter explains evaluations in detail. Here is a list of key points:

• An evaluation allows you to measure your model, ensemble, logistic regression, deepnet, and fusion performance.

• In BigML you can perform two types of evaluations: single evaluations and cross-validation evaluations.

• You need a model and a testing dataset to create a single evaluation. (See Figure 7.115 .)

• You just need an existing dataset to create a cross-validation evaluation. (See Figure 7.116 .)

• BigML provides you a range of configuration options before creating your evaluation.

• Performance measures are different for classification and regression models.

• The confusion matrix is a key element to evaluate the performance of classification models.

• You can compare your evaluations measures against models using the mean, the mode, and a random value to predict.

• BigML provides different visualizations for the ROC curve, the Precision-Recall curve, the Gain curve, and the Lift curve along with their AUC, K-S statistic and other metrics.

• You can compare two or more evaluations built with different configurations and algorithms to select the model with the best performance.

• You can download your confusion matrix in Excel format.

• You can create and use evaluations via the BigML API and bindings.

• You can add descriptive information to your evaluations.

• You can move your evaluations between projects.

• You can share your evaluations with other people using the secret link.

• You can stop your evaluations creation by deleting them.

• You can permanently delete an existing evaluation.