Histogram Gallery

This gallery page showcases histogram-based plots from k-diagram designed for visualizing the distribution of one-dimensional data, which is fundamental for evaluating forecast errors and uncertainty.

Note

You need to run the code snippets locally to generate the plot images referenced below. Ensure the image paths in the .. image:: directives match where you save the plots.

Histogram with KDE Overlay

Uses plot_hist_kde() to visualize the distribution of a variable. It combines a traditional histogram with a smooth Kernel Density Estimate (KDE) curve to provide a detailed view of the data’s shape, central tendency, and spread.

import kdiagram.utils.hist as kdh
import numpy as np
import matplotlib.pyplot as plt

# --- Data Generation ---
# Simulate sensor readings with a normal distribution
np.random.seed(42)
sensor_readings = np.random.normal(loc=100, scale=15, size=1000)

# --- Plotting ---
kdh.plot_hist_kde(
    sensor_readings,
    bins=40,
    title="Distribution of Sensor Readings",
    x_label="Temperature (°C)",
    kde_color="#FF5733", # Custom reddish-orange
    # Save the plot (adjust path relative to docs/source/)
    savefig="gallery/images/gallery_utils_sensor_reading.png"
)
plt.close() # Close plot after saving

🧠 Analysis and Interpretation

The Histogram with KDE Overlay is a classic and effective tool for inspecting the distribution of any continuous variable, such as prediction errors or interval widths.

Analysis and Interpretation:

• Histogram (Blue Bars): The height of each bar shows the frequency (or density) of data points falling within that specific bin or range. It gives a discretized view of the distribution.

• KDE Curve (Orange Line): The Kernel Density Estimate provides a smooth, continuous line that estimates the underlying probability density function of the data. It helps to identify the shape, peaks (modes), and skewness of the distribution more clearly than the histogram alone.

• Combined View: Together, they confirm the data’s characteristics. The KDE should generally follow the contour of the histogram bars.

🔍 Key Insights from this Example:

• The data (simulated sensor readings) appears to be normally distributed, which is confirmed by the classic “bell curve” shape of both the histogram and the KDE.

• The central peak is located around 100, indicating that this is the most frequent sensor reading (the mean or mode).

• The distribution is largely symmetrical, with tails tapering off evenly on both sides of the peak.

💡 When to Use:

• Error Analysis: To check if your model’s prediction errors (actual - predicted) are normally distributed and centered at zero, which is a common assumption and sign of a well-calibrated model.

• Feature Distribution: To understand the distribution of input features before modeling.

• Uncertainty Characterization: To analyze the distribution of prediction interval widths. A narrow, single-peaked distribution is often desirable, indicating consistent uncertainty estimates.