Multisite data generation

Imports

import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns

from uniharmony import verbosity
from uniharmony.datasets import make_multisite_classification


sns.set_theme(style="whitegrid")
verbosity("warning")

Data generation

# We will start with the function as default
X, y, sites = make_multisite_classification()
df = pd.DataFrame({"Class": y, "Site": sites})
print(f"X has {X.shape[0]} examples and {X.shape[1]} features")

plt.figure(figsize=[10, 6])
plt.title("Class and site distribution")
sns.countplot(df, x="Class", hue="Site")
plt.grid(axis="y", color="black", alpha=0.5, linestyle="--")

X has 1000 examples and 10 features

# Let's increase the number of sites.
# Note that the total number of samples is the same, but the distribution changed.
fig, axes = plt.subplots(1, 3, figsize=[15, 5])

for idx, n_sites in enumerate([2, 4, 10]):
    X, y, sites = make_multisite_classification(n_sites=n_sites)
    print(f"For n_sites {n_sites}, X has {X.shape[0]} examples and {X.shape[1]} features")

    df_plot = pd.DataFrame(
        {
            "Class": y,
            "Site": sites,
            "Feature 1": X[:, 0],
            "Feature 2": X[:, 1],
        }
    )
    sns.countplot(df_plot, x="Class", hue="Site", ax=axes[idx])
    plt.title("Class and site distribution")

    axes[idx].set_title(f"Class and site distribution with {n_sites} sites")
    axes[idx].grid(axis="y", color="black", alpha=0.5, linestyle="--")

plt.tight_layout()

For n_sites 2, X has 1000 examples and 10 features
For n_sites 4, X has 1000 examples and 10 features
For n_sites 10, X has 1000 examples and 10 features

Plotting

fig, axes = plt.subplots(1, 3, figsize=[15, 5])

for idx, n_classes in enumerate([2, 4, 10]):
    X, y, sites = make_multisite_classification(n_classes=n_classes)
    print(f"For n_classes {n_classes}, X has {X.shape[0]} examples and {X.shape[1]} features")

    df_plot = pd.DataFrame(
        {
            "Class": y,
            "Site": sites,
            "Feature 1": X[:, 0],
            "Feature 2": X[:, 1],
        }
    )
    sns.countplot(df_plot, x="Class", hue="Site", ax=axes[idx])
    axes[idx].set_title(f"Class and site distribution with {n_classes} classes")
    axes[idx].grid(axis="y", color="black", alpha=0.5, linestyle="--")

plt.tight_layout()

For n_classes 2, X has 1000 examples and 10 features
For n_classes 4, X has 1000 examples and 10 features
For n_classes 10, X has 1000 examples and 10 features

Variations

Changing the class balance

As default, the created problem is balanced accross sites, so all classes has the same number of examples in each site. We can control that using the balance_per_site argument of the function.

# In a binary classification problem, you need to set which
# proportion of class 1 is present in that site.
# Thus, the number of elements in the list must match the number of sites.
fig, axes = plt.subplots(1, 3, figsize=[15, 5])
balance_per_site_list = [[0.5, 0.5], [0.3, 0.7], [0.1, 0.9]]
for idx, balance_per_site in enumerate(balance_per_site_list):
    X, y, sites = make_multisite_classification(balance_per_site=balance_per_site)
    print(f"For n_classes {n_classes}, X has {X.shape[0]} examples and {X.shape[1]} features")

    df_plot = pd.DataFrame(
        {
            "Class": y,
            "Site": sites,
            "Feature 1": X[:, 0],
            "Feature 2": X[:, 1],
        }
    )
    sns.countplot(df_plot, x="Class", hue="Site", ax=axes[idx])
    axes[idx].set_title(f"Balance per classes {balance_per_site} ")
    axes[idx].grid(axis="y", color="black", alpha=0.5, linestyle="--")

plt.tight_layout()


# Now that we changed the balance, we have different occurence in different sites. Note that the total number of classes is still balanced, is the distribution accross sites that changed.

For n_classes 10, X has 1000 examples and 10 features
For n_classes 10, X has 1000 examples and 10 features
For n_classes 10, X has 1000 examples and 10 features

Changing the signal type

The signal type is used to generate a base problem, using sklearn functions. From this generated problem, uniharmony adds an Effect of Site and noise on top.

fig, axes = plt.subplots(1, 5, figsize=[20, 7])

for idx, signal_type in enumerate(["linear", "moons", "circles", "blobs", "make_gaussian_quantiles"]):
    X, y, sites = make_multisite_classification(
        n_features=2,
        signal_type=signal_type,
        noise_strength=0.1,
        site_effect_strength=0,
    )

    df_plot = pd.DataFrame(
        {
            "Class": y,
            "Site": sites,
            "Feature 1": X[:, 0],
            "Feature 2": X[:, 1],
        }
    )

    sns.scatterplot(df_plot, x="Feature 1", y="Feature 2", hue="Class", ax=axes[idx])
    axes[idx].set_title(f"Signal Type: {signal_type}")
    axes[idx].grid(axis="y", color="black", alpha=0.5, linestyle="--")

plt.tight_layout()

fig, axes = plt.subplots(1, 3, figsize=[15, 5])

for idx, signal_st in enumerate([0, 1, 5]):
    X, y, sites = make_multisite_classification(
        n_features=2,
        signal_strength=signal_st,
        noise_strength=0.1,
        site_effect_strength=0,
    )

    df_plot = pd.DataFrame(
        {
            "Class": y,
            "Site": sites,
            "Feature 1": X[:, 0],
            "Feature 2": X[:, 1],
        }
    )

    sns.scatterplot(df_plot, x="Feature 1", y="Feature 2", hue="Class", ax=axes[idx])
    axes[idx].set_title(f"Signal Strength: {signal_st}")
    axes[idx].grid(axis="y", color="black", alpha=0.5, linestyle="--")

plt.tight_layout()

2026-05-18 13:04:36 [warning  ] signal_strength is 0. Adding a delta (1e-6) to signal_strength to avoid degenerate data.

Effect of Site effect

fig, axes = plt.subplots(1, 3, figsize=[15, 5])

for idx, site_st in enumerate([0, 1, 5]):
    X, y, sites = make_multisite_classification(
        n_sites=4,
        n_features=2,
        signal_strength=0.001,
        noise_strength=0,
        site_effect_strength=site_st,
        signal_type="circles",
        site_effect_homogeneous=True,
    )

    df_plot = pd.DataFrame(
        {
            "Class": y,
            "Site": sites,
            "Feature 1": X[:, 0],
            "Feature 2": X[:, 1],
        }
    )

    sns.scatterplot(df_plot, x="Feature 1", y="Feature 2", hue="Site", ax=axes[idx])
    axes[idx].set_title(f"Effect of Site Strength: {site_st}")
    axes[idx].grid(axis="y", color="black", alpha=0.5, linestyle="--")

plt.tight_layout()

Site effect homogeneous

fig, axes = plt.subplots(1, 3, figsize=[15, 5])

for idx, site_st in enumerate([0, 1, 5]):
    X, y, sites = make_multisite_classification(
        n_sites=10,
        n_features=2,
        signal_strength=0.001,
        noise_strength=0,
        site_effect_strength=site_st,
        signal_type="circles",
        noise=0.1,
        site_effect_homogeneous=True,
    )

    df_plot = pd.DataFrame(
        {
            "Class": y,
            "Site": sites,
            "Feature 1": X[:, 0],
            "Feature 2": X[:, 1],
        }
    )

    sns.scatterplot(df_plot, x="Feature 1", y="Feature 2", hue="Class", ax=axes[idx])
    axes[idx].set_title(f"Effect of Site Strength: {site_st}")
    axes[idx].grid(axis="y", color="black", alpha=0.5, linestyle="--")

plt.tight_layout()
fig, axes = plt.subplots(1, 3, figsize=[15, 5])

for idx, site_st in enumerate([0, 1, 5]):
    X, y, sites = make_multisite_classification(
        n_sites=10,
        n_features=2,
        signal_strength=0.001,
        noise_strength=0,
        site_effect_strength=site_st,
        signal_type="circles",
        noise=0.1,
        site_effect_homogeneous=False,
    )

    df_plot = pd.DataFrame(
        {
            "Class": y,
            "Site": sites,
            "Feature 1": X[:, 0],
            "Feature 2": X[:, 1],
        }
    )

    sns.scatterplot(df_plot, x="Feature 1", y="Feature 2", hue="Class", ax=axes[idx])
    axes[idx].set_title(f"Effect of Site Strength: {site_st}")
    axes[idx].grid(axis="y", color="black", alpha=0.5, linestyle="--")
plt.tight_layout()

• 
• 

Noise effect

verbosity("warning")

fig, axes = plt.subplots(1, 3, figsize=[15, 5])

for idx, noise_st in enumerate([0.01, 0.1, 0.5]):
    X, y, sites = make_multisite_classification(
        n_features=2,
        signal_strength=1,
        noise_strength=noise_st,
        site_effect_strength=0,
    )

    df_plot = pd.DataFrame(
        {
            "Class": y,
            "Site": sites,
            "Feature 1": X[:, 0],
            "Feature 2": X[:, 1],
        }
    )

    sns.scatterplot(df_plot, x="Feature 1", y="Feature 2", hue="Class", ax=axes[idx])
    axes[idx].set_title(f"Noise Strength: {noise_st}")
    axes[idx].grid(axis="y", color="black", alpha=0.5, linestyle="--")
    axes[idx].set_xlim(-5, 5)
    axes[idx].set_ylim(-5, 5)

plt.tight_layout()

fig, axes = plt.subplots(1, 3, figsize=[15, 5])

for idx, strength in enumerate([0, 1, 50]):
    X, y, sites = make_multisite_classification(
        n_sites=3, n_features=2, signal_strength=4, site_effect_strength=strength, signal_type="blobs", random_state=23
    )

    df_plot = pd.DataFrame(
        {
            "Class": y,
            "Site": sites,
            "Feature 1": X[:, 0],
            "Feature 2": X[:, 1],
        }
    )

    sns.scatterplot(df_plot, x="Feature 1", y="Feature 2", hue="Site", ax=axes[idx])
    axes[idx].set_title(f"Signal Strength: {strength}")
    axes[idx].grid(axis="y", color="black", alpha=0.5, linestyle="--")


plt.tight_layout()