Provide labels
CSV with y_true and y_pred columns, or two pasted columns. Labels can be text or numbers.
Fine-tuning
Classification Metrics & Confusion Matrix
Evaluation without a notebook. Paste two columns or drop a CSV with true and predicted labels (and optionally probabilities), and the tool prints the confusion matrix, per-class precision, recall and F1, macro and weighted averages, Cohen's kappa and, with probabilities, ROC-AUC for binary tasks.
Loading the tool…
How it works
Read the matrix
Rows are true classes, columns predicted; the heat colouring points at the confusions that matter.
Export
Metrics as JSON or Markdown table; the matrix as SVG or CSV.
Formats, limits and the numbers that matter
Averages
Macro treats classes equally; weighted weights by support; micro equals accuracy for single-label tasks. Report macro when classes are imbalanced.
Thresholds
With probabilities for a binary task, the threshold slider recomputes everything and draws the precision-recall curve.
Multi-label
Semicolon-separated labels per row are treated as multi-label; metrics are per label with micro and macro averages.
Common questions
How many examples for a reliable F1?
The 95 percent interval on a proportion at n=100 is about ±10 points; at n=1,000 about ±3. The tool prints a Wilson interval for accuracy.
Detection metrics?
Use the mAP Calculator for boxes.
Can I compare two models?
Add a second prediction column and both are reported side by side.