Task Metrics#
Kolena supports automatic calculation of a number of metrics commonly used in machine learning tasks.
Regression#
For regression problems, Kolena web application currently supports mean_absolute_error,
mean_squared_error,
root_mean_squared_error, r^2,
pearson_correlation and spearman_correlation
Follow the steps below to setup Regression metrics:
Binary Classification#
For binary classification problems, Kolena web application currently supports accuracy,precision,
recall and f1_score.
Follow the steps below to setup Binary Classification metrics:
Multiclass Classification#
For multiclass classification problems, Kolena web application currently supports accuracy,precision,
recall and f1_score.
You are able to apply macro,
micro and
weighted averaging methods to above metrics.
Object detection#
For object detection problems, Kolena web application currently
supports average precision, precision,
recall and f1_score.
Required fields
Kolena attempts to automatically detect fields for true positive, false positive and false negative counts. For more information on how to prepare your object detection data, please visit Formatting results for Object Detection
Thresholded Object Metrics#
Kolena provides the flexibility to calculate and use metrics that are threshold dependent. For more information about how to setup thresholded results checkout the Thresholded Results page.
Once your can uploaded thresholded results, the option to setup thresholded metrics will be available.
Custom Metrics#
Kolena provides out-of-the-box aggregation options for your datapoint level evaluations that
correspond with your desired metrics. For numeric evaluations you are able to
select from count, mean, median, min, max, stddev and sum aggregations options.
For categorical evaluations (class label, boolean, etc) rate and count aggregation options are available.
The Kolena web application currently supports precision,
recall, f1_score,
accuracy, false_positive_rate,
and true_negative_rate.
To leverage these, add the following columns to your model result file: count_TP, count_FP, count_FN, count_TN.
