analytics~metrics.PredictionCurve

Example

// import metrics module
var metrics = require('qminer').analytics.metrics;

// true and predicted lables
var true_lables = [0, 1, 0, 0, 1];
var pred_prob = [0.3, 0.5, 0.2, 0.5, 0.8];

// create predictionCurve instance
var predictionCurve = new metrics.PredictionCurve();

// simulate data flow
for (var i in true_lables) {
   // push new value
   predictionCurve.push(true_lables[i], pred_prob[i]);
}

var roc = predictionCurve.roc(); // get ROC
var auc = predictionCurve.auc(); // get AUC
var pr = predictionCurve.precisionRecallCurve() // get precision-recall curve

Parameters

allNegatives  number

Count of all negative examples.

allPositives  number

Count of all positive examples.

grounds  module:la.Vector

Store of ground truths.

length  number

Count of all examples.

predictions  module:la.Vector

Store of predictions.

auc([sample]) → number

Get Area Under the Curve (AUC) of the current curve.

Returns

numberB Area under ROC curve.

bestF1() → number

Gets threshold for prediction score, which results in the highest F1.

Returns

numberB Threshold with highest F1 score.

breakEvenPoint() → number

Get break-even point, the value where precision and recall intersect.

Returns

numberB Break-even point.

desiredPrecision(desiredPrecision) → number

Gets threshold for prediction score, nearest to specified precision.

Returns

numberB Threshold for prediction score, nearest to specified precision.

desiredRecall(desiredRecall) → number

Gets threshold for prediction score, nearest to specified recall.

Returns

numberB Recal Score Threshold. Threshold for recall score, nearest to specified recall.

precisionRecallCurve([sample]) → module:la.Matrix

Get precision recall curve sampled on sample points.

Returns

module:la.MatrixB Precision-recall pairs.

push(ground, predicted)

Add new measurement with ground score (1 or -1) and predicted value or integer array (when there are zero or more then one lables).

roc([sample]) → module:la.Matrix

Get Receiver Operating Characteristic (ROC) parametrization sampled on sample points.

Returns

module:la.MatrixB A matrix with increasing false and true positive rates.