MLClassificationEvaluation
Classification detail of a holdout evaluation
Fields
For a binary model, the class counted as positive: the rarer class on the training side
Class labels, in the order of the confusion matrix
Binary confusion counts at thresholds 0.05 to 0.95
Reliability diagram of the positive-class probability
Confusion matrix: rows are actual classes, columns predicted, both in classes order; null above 50 classes
Whether the confusion matrix was omitted because there were too many classes
Per-class precision, recall and F1
Held-out rows whose class never appeared in training; counted as wrong
Held-out rows skipped because the model produced a non-finite probability
Referenced by (1)
Exact type SDL
"""Classification detail of a holdout evaluation"""
type MLClassificationEvaluation {
"""
For a binary model, the class counted as positive: the rarer class on the training side
"""
positiveClass: String
"""Class labels, in the order of the confusion matrix"""
classes: [String!]!
"""Binary confusion counts at thresholds 0.05 to 0.95"""
thresholdCurve: [MLThresholdPoint!]!
"""Reliability diagram of the positive-class probability"""
calibration: [MLCalibrationBin!]!
"""
Confusion matrix: rows are actual classes, columns predicted, both in classes order; null above 50 classes
"""
confusion: [[Int!]!]
"""
Whether the confusion matrix was omitted because there were too many classes
"""
confusionTruncated: Boolean!
"""Per-class precision, recall and F1"""
perClass: [MLClassMetrics!]!
"""Held-out rows whose class never appeared in training; counted as wrong"""
unseenClassRows: Int!
"""
Held-out rows skipped because the model produced a non-finite probability
"""
nonFiniteRows: Int!
}