Characterise a multisite problem

The first step before applying any harmonization technique is to understand and characterize our data

Imports

import matplotlib.pyplot as plt
import seaborn as sns

from uniharmony import verbosity
from uniharmony.datasets import (
    get_multisite_data_statistics,
    make_multisite_classification,
    print_statistics_summary,
)
from uniharmony.plot import plot_features_by_site


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

Data generation

Let’s use the multisite data generator to simulate some data

print("Generating example data...")
X, y, sites = make_multisite_classification(
    n_sites=5,
    n_samples=1000,
    n_features=10,
    n_classes=3,
    random_state=42,
)

print("\n" + "=" * 60)

Generating example data...

============================================================

Now let’s compute some statistics

print("Computing statistics...")
print("=" * 60)

# Compute statistics
stats = get_multisite_data_statistics(
    X=X,
    y=y,
    sites=sites,
    feature_names=[f"feat_{i}" for i in range(X.shape[1])],
)
verbosity("info")
# Print summary
print_statistics_summary(stats)
verbosity("warning")

Computing statistics...
============================================================
2026-05-18 13:03:34 [info     ] ============================================================
2026-05-18 13:03:34 [info     ] DATASET STATISTICS SUMMARY
2026-05-18 13:03:34 [info     ] ============================================================
2026-05-18 13:03:34 [info     ]
OVERALL:
2026-05-18 13:03:34 [info     ]   Samples: 1000
2026-05-18 13:03:34 [info     ]   Features: 10
2026-05-18 13:03:34 [info     ]   Sites: 5
2026-05-18 13:03:34 [info     ]   Classes: 3
2026-05-18 13:03:34 [info     ]
CLASS DISTRIBUTION:
2026-05-18 13:03:34 [info     ]   class_0: 335 samples (33.5%)
2026-05-18 13:03:34 [info     ]   class_1: 335 samples (33.5%)
2026-05-18 13:03:34 [info     ]   class_2: 330 samples (33.0%)
2026-05-18 13:03:34 [info     ]
SITE DISTRIBUTION:
2026-05-18 13:03:34 [info     ]   site_0: 200 samples (20.0%)
2026-05-18 13:03:34 [info     ]   site_1: 200 samples (20.0%)
2026-05-18 13:03:34 [info     ]   site_2: 200 samples (20.0%)
2026-05-18 13:03:34 [info     ]   site_3: 200 samples (20.0%)
2026-05-18 13:03:34 [info     ]   site_4: 200 samples (20.0%)
2026-05-18 13:03:34 [info     ]
SITE STATISTICS (summary):
2026-05-18 13:03:34 [info     ]   site_0:
2026-05-18 13:03:34 [info     ]     Samples: 200
2026-05-18 13:03:34 [info     ]     Class distribution: {'class_0': 67, 'class_1': 67, 'class_2': 66}
2026-05-18 13:03:34 [info     ]   site_1:
2026-05-18 13:03:34 [info     ]     Samples: 200
2026-05-18 13:03:34 [info     ]     Class distribution: {'class_0': 67, 'class_1': 67, 'class_2': 66}
2026-05-18 13:03:34 [info     ]   site_2:
2026-05-18 13:03:34 [info     ]     Samples: 200
2026-05-18 13:03:34 [info     ]     Class distribution: {'class_0': 67, 'class_1': 67, 'class_2': 66}
2026-05-18 13:03:34 [info     ]   site_3:
2026-05-18 13:03:34 [info     ]     Samples: 200
2026-05-18 13:03:34 [info     ]     Class distribution: {'class_0': 67, 'class_1': 67, 'class_2': 66}
2026-05-18 13:03:34 [info     ]   site_4:
2026-05-18 13:03:34 [info     ]     Samples: 200
2026-05-18 13:03:34 [info     ]     Class distribution: {'class_0': 67, 'class_1': 67, 'class_2': 66}
2026-05-18 13:03:34 [info     ]
FEATURE STATISTICS (first 5 features):
2026-05-18 13:03:34 [info     ]   feat_0: mean=0.8137, std=1.6804, MAD=1.2405
2026-05-18 13:03:34 [info     ]   feat_1: mean=-0.2313, std=1.9461, MAD=1.4023
2026-05-18 13:03:34 [info     ]   feat_2: mean=1.0819, std=2.0994, MAD=1.4588
2026-05-18 13:03:34 [info     ]   feat_3: mean=0.8475, std=1.6420, MAD=1.2095
2026-05-18 13:03:34 [info     ]   feat_4: mean=0.7886, std=1.5675, MAD=1.1310
2026-05-18 13:03:34 [info     ]
CORRELATIONS:
2026-05-18 13:03:34 [info     ]   Average Inter-Site Correlation: 0.9485
2026-05-18 13:03:34 [info     ] ============================================================

# Same plot individual points overlay
fig2, ax2 = plot_features_by_site(
    X,
    sites,
    figsize=(14, 7),
    rotation=45,
    show_points=True,
    title="All Features by Site (with individual points)",
)
plt.show()