uncomplicate.diamond.metrics — org.uncomplicate/deep-diamond-base 0.34.0

Classification evaluation functions.

classification-metrics clj

(classification-metrics cm)

(classification-metrics real pred)

(classification-metrics real pred n)

Computes most common classification metrics from confusion matrix cm (which can be provided, or computed internally from real and prod). The result is a Clojure map, with keys :metrics and :macro. :metrics are returned as a map of neanderthal vectors: :accuracy, :f1, sensitivity, :specificity, etc. The :macro key contains rates from cm (true positive rate, true negative rate, etc.).

Arguments:

• cm: pre-computed confusion matrix.
• real: values as they are supposed to be.
• pred: values predicted by an algorithm.
• n: number of categories in real and pred.

Please see relevant Wikipedia article for the quick reference.

confusion-matrix clj

(confusion-matrix real pred)

(confusion-matrix real pred n)

Returns the confusion matrix between vectors, matrices, or tensors with real and pred (predicted) categories. For example, a typical confusion matrix with two categories, shows the number of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN). This function generalizes this to n categories, where there are n-1 different ways to be wrong, that is, when the predicted category is not the same as the real category, we can treat these mistakes per each category, instead in bulk. The categories are natural numbers in theory, and this function can process both positive integers and floats. If the data in real or prod floats, their whole value part will be consider to be the category index (for example, both 4.0 and 4.889 fall into category 4).

Arguments:

• real: values as they are supposed to be.
• pred: values predicted by an algorithm.
• n: number of categories in real and pred.

contingency-totals clj

(contingency-totals cm)

Calculates true positives, false positives, and false negatives from the confusion matrix cm. Returns a 3 x n matrix, where columns are categories, and rows are TP, FP, and FN. Since multi-category confusion matrix is per-category, there is no point in making distinction between true positive and true negative; both are 'true this', whatever 'this' may be (this category is 'positive', and all other categories are 'negative').