Analyze feature distributions using flexible logarithmic conversion controls. Review transformed values, statistics, and exportable reports. Build cleaner training datasets with confidence and visual clarity.
Enter values separated by commas, spaces, semicolons, or new lines.
This example uses log1p with no extra scaling.
| Index | Original Value | Adjusted Input | Transformed Value | Final Output |
|---|---|---|---|---|
| 1 | 0.0000 | 0.0000 | 0.0000 | 0.0000 |
| 2 | 1.0000 | 1.0000 | 0.6931 | 0.6931 |
| 3 | 3.0000 | 3.0000 | 1.3863 | 1.3863 |
| 4 | 7.0000 | 7.0000 | 2.0794 | 2.0794 |
| 5 | 15.0000 | 15.0000 | 2.7726 | 2.7726 |
Log transformation compresses large values while preserving order. It helps reduce right skew, stabilize variance, and make feature distributions easier for models to learn.
General adjusted input: x_adjusted = x + pseudocount + effective shift
Natural log: y = ln(x_adjusted)
Base-10 log: y = log10(x_adjusted)
Base-2 log: y = log2(x_adjusted)
Custom base: y = ln(x_adjusted) / ln(base)
Log1p: y = ln(1 + x_adjusted)
Z-score scaling: z = (y - mean(y)) / std(y)
Min-max scaling: m = (y - min(y)) / (max(y) - min(y))
The calculator also reports count, mean, median, variance, standard deviation, and skewness before and after transformation.
It reduces heavy right skew, compresses large outliers, and often makes numeric features easier for algorithms to model. This can improve stability and interpretability during feature engineering.
Standard log needs strictly positive values. Log1p computes ln(1 + x), so zero values remain valid without needing an added constant in many common workflows.
Use a pseudocount when zeros would otherwise break the transform. Small constants are common in count-based features, sparse frequency data, and positive measurements with many zero entries.
Shift moves every value upward before transformation. It is useful when your dataset includes negatives or very small values that would fall outside the valid log domain.
That depends on the model. Distance-based and gradient-based methods often benefit from scaling after log transformation, while tree-based models usually need it less.
Yes. Log transformations are monotonic for valid positive inputs, so relative ordering stays consistent even though numeric spacing becomes compressed.
Skewness measures asymmetry in the distribution. Comparing raw and transformed skewness helps you check whether the log step made the feature more balanced.
Yes. Base e, 10, and 2 are common, but custom bases can match domain conventions. The main effect is output scale, not rank ordering.
Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.