Feature Scaling Calculator

Scale each numeric column independently. Load/paste CSV or use Demo, then choose a transform above each column. Edit original values to change the source; uncheck it to see the current result. Unchanged columns may contain text. CSV export saves the currently displayed view with full numeric precision; the table displays six significant digits.

Min–Max: (x−min)/(max−min), mapping the observed extrema to 0 and 1. Z-score: (x−mean)/s. Choose n−1 for sample standard deviation or n for population standard deviation, as in scikit-learn StandardScaler. Constant columns become 0 with scale 1, including a single value. Log₁₀ requires strictly positive values; it is a logarithmic transform, not standardization. Demo gives A:0/0.5/1, B:−1/0/1 (n−1), C:0/1/2.

Automatic header detection treats an entirely nonnumeric first row as headers; choose Headers or Data explicitly if needed. Delimiters are detected; decimal commas work in quoted cells or semicolon-separated CSV. Empty or invalid numeric cells are errors, never zero. Limits: 200 data rows, 20 columns, 1 MiB, numeric magnitude at most 10¹⁵⁰. Oversized imports are rejected, preserving the previous table. Everything runs locally.

Parameters are fitted on all imported rows. For machine learning fit on training data only, then reuse those parameters on test data to prevent leakage. This tool does not separate training/test sets or apply saved fits to another dataset. Scaling is sensitive to outliers; it does not make arbitrary distributions normal. Min–Max inverse is y·range+min; Z-score inverse is y·s+mean; log inverse is 10ʸ.

Only input values are remembered in this browser. Calculated results and experiment histories are not saved.