Visualization

🧭 Purpose

This page explains the visualization outputs used in Sake to inspect data quality, understand distributions, evaluate models, and communicate results.

Visual review is an important complement to statistical metrics. A model can report acceptable summary metrics while still showing bias, outliers, unstable residuals, or poor performance for important classes.

📊 Visualization Categories

Category	Purpose
Data profile charts	Inspect distributions, missing values, and outliers.
Statistical charts	Compare groups and relationships.
Classification diagnostics	Review categorical prediction performance.
Regression diagnostics	Review numeric prediction performance.
Feature charts	Identify influential predictors or reduced-dimension structure.
Benchmark charts	Compare model performance across algorithms.

📈 Distribution Plots

Histograms show the distribution of numeric fields.

import matplotlib.pyplot as plt

column = "obligations"

plt.figure()
df_accounts[column].dropna().hist(bins=30)
plt.xlabel(column)
plt.ylabel("Count")
plt.title(f"Distribution of {column}")
plt.show()

Use distribution plots to identify skew, zeros, negative values, and extreme values.

📦 Boxplots

Boxplots help identify outliers and compare groups.

column = "obligations"

plt.figure()
df_accounts.boxplot(column=column)
plt.title(f"Boxplot of {column}")
plt.show()

Grouped boxplot:

df_accounts.boxplot(column="obligations", by="account_type")
plt.title("Obligations by Account Type")
plt.suptitle("")
plt.show()

🔥 Correlation Heatmap

import matplotlib.pyplot as plt

df_corr = df_accounts.select_dtypes(include="number").corr()

plt.figure()
plt.imshow(df_corr)
plt.xticks(range(len(df_corr.columns)), df_corr.columns, rotation=90)
plt.yticks(range(len(df_corr.columns)), df_corr.columns)
plt.colorbar()
plt.title("Correlation Matrix")
plt.tight_layout()
plt.show()

✅ Classification Visuals

Visual	Use
Confusion matrix	Shows correct and incorrect class predictions.
ROC curve	Shows tradeoff between true-positive and false-positive rates.
Precision-recall curve	Useful when positive classes are rare.
Class distribution chart	Shows class imbalance.
Feature importance chart	Shows influential predictors for supported models.

Confusion Matrix Example

from sklearn.metrics import ConfusionMatrixDisplay

ConfusionMatrixDisplay.from_predictions(y_test, predictions)
plt.title("Confusion Matrix")
plt.show()

📉 Regression Visuals

Visual	Use
Predicted versus actual	Shows fit and systematic bias.
Residuals versus predicted	Shows error structure.
Residual histogram	Shows error distribution.
Feature importance	Shows influential predictors.
Benchmark bar chart	Compares model metrics.

Predicted Versus Actual Example

plt.figure()
plt.scatter(y_test, predictions)
plt.xlabel("Actual")
plt.ylabel("Predicted")
plt.title("Predicted Versus Actual")
plt.show()

Residual Histogram Example

residuals = y_test - predictions

plt.figure()
residuals.hist(bins=30)
plt.xlabel("Residual")
plt.ylabel("Count")
plt.title("Residual Distribution")
plt.show()

🧮 Model Benchmark Chart

df_results.plot(
    kind="bar",
    x="model",
    y="score",
    legend=False
)

plt.ylabel("Score")
plt.title("Model Benchmark")
plt.tight_layout()
plt.show()

🏛️ Budget Execution Visualization Notes

Budget data often includes very large outliers. Consider:

• log-scale views for high-dollar variables
• separate views by account type
• showing top accounts by obligation or outlay
• reviewing negative values separately
• labeling charts clearly with fiscal year or reporting period
• avoiding misleading axes for financial data

✅ Visualization Checklist

Before using charts in reports:

• Chart title describes the measure and population.
• Axes are labeled.
• Units are clear.
• Outliers are not hidden without explanation.
• Group definitions are documented.
• Model diagnostics support the metric interpretation.
• The chart is readable in dark-mode documentation.