Understand feature distributions
Histograms are useful for visually inspecting feature distributions, and especially to check if it follows a normal (Gaussian) distribution.
Normal distributions provide several useful properties for statistical modelling and is one of the key assumptions of linear regression.
Selecting a feature
Once you load a dataset in the Datasets tab. You can start visualising the distribution of your features by moving on to the Histogram tab.
Following from the Datasets tutorial, we use the Iris dataset as an example.
Numerical features
When you select a Numerical feature, values are counted and put into small intervals (bins).
Visually, this means that you can use the X-axis to see where those intervals fall and the Y-axis indicates the count of data points within each bin. For instance, if a bin for values between 1.4 and 1.6 has a count of 26, then there were 26 data points which fall into the range [1.4, 1.6).
This example is for the PetalLengthCm feature from the Iris dataset, which measures a flower's petal length in centimeters.
Categorical features
When you select a Categorical feature, value counts are exact and the Y-axis shows how many data points had the exact values shown on the X-axis.
From the Iris dataset, the histogram for the Species feature shows how many data points correspond to each species. In this example, you can see that each species is represented equally in this dataset.
Feature Transformation
Feature transformation methods
Visprex provides three feature transformations out-of-the-box, squared, log10, and natural log. You can see this in the f(x) section below the list of features.
x: Default value. No transformation.x²: Squared value.log10(x): Logarithmic transformation with base 10.ln(x): Logarithmic transformation with base e.
When is log transformation useful?
Log transformation is useful when the distribution of your selected feature is heavily skewed to either side.
For example, most distributions around money (income, house properties, GDP) tends to have a skewed distribution where a small group of data points show extremely high values.
Choose world-dataset-2023.csv from example datasets.
Click on Histogram tab and select the GDP feature. We can see that the distribution is heavily skewed to the right where countries with higest GDP values lie on the far right side of the histrogram.
Click on log10(x) to transform the values to the base 10 logarithmic scale, then you can see that the distribution now closely approximates the normal distribution.
Given that this value is denominated by the US Dollars and the base is 10, you can now read the bins on this histogram as the number of countries that fall into the range between 10^(x) and 10^(x+d), where d is the difference between the start of the interval and the end of the interval on the histogram.
