import numpy as np
rng = np.random.RandomState(42)

n_samples = 100
neg_samples = rng.rand(n_samples // 2, 2)
pos_samples = rng.rand(n_samples // 2, 2) + np.array([3, 3])

X = np.vstack([neg_samples, pos_samples])
y = np.concatenate([-np.ones(n_samples // 2), np.ones(n_samples // 2)])

from sklearn.svm import SVC
svc = SVC(kernel="linear")

svc.fit(X, y)

svc.dual_coef_ == np.array([0.1999464, -0.1999464])

w = svc.dual_coef_.dot(svc.support_vectors_)
margin = 2 / np.sqrt((w ** 2).sum())

from sklearn.metrics import euclidean_distances
distances = euclidean_distances(neg_samples, pos_samples)

assert np.abs(margin - distances.min()) < 1e-6