confusion

Docs: GoDoc

Confusion implements a confusion matrix: records output responses for discrete categories / classes.

• Rows (outer dimension) are for each class as the ground truth, correct answer.

• Columns (inner dimension) are the response generated for each ground-truth class.

The main result is in the Prob field, computed from the Sum and N values added incrementally.

Main API:

• InitFromLabels to initialize with list of class labels and display font size.
• Incr on each trial with network's response index and correct target index.
• Probs when done, to compute probabilities from accumulated data.
• SaveCSV / OpenCSV for saving / loading data (for nogui usage).

The TFPN matrix keeps a record of true/false positives (tp/fp) and true/false negatives (tn/fn) for each category/class. This table is used to calculate F1 scores either by class or across classes

A beginner’s guide on how to calculate Precision, Recall, F1-score for a multi-class classification problem can be found at https://towardsdatascience.com/confusion-matrix-for-your-multi-class-machine-learning-model-ff9aa3bf7826

API:

• SumTFPN to calculate the tp, fp, fn and tn scores for each class
• ScoreClass calculates the precision and recall scores that are needed for the F1 score
• ScoreMatrix uses the values calculated by ScoreClass to generate 3 different F1 scores for the entire matrix
  • F1 Micro
  • F1 Macro
  • F1 Weighted