3D Logistic Regression Trainer

The same batch gradient descent that trains a 2-D classifier extends directly to a real 3-feature space, and the result is genuinely 3-D geometry: a separating plane, not a line. This simulator scatters two overlapping Gaussian blobs of points through (x₁, x₂, x₃) space, then repeatedly computes the logistic regression model's actual predicted probability for every point, measures the real binary cross-entropy loss, and steps the three weights and bias down their true gradient — rebuilding the decision plane from those weights every epoch so you watch it rotate and slide into place as loss falls. Orbit the camera to see the plane's tilt from any angle, and tune learning rate, L2 regularization, class separation and noise to explore how each shapes convergence.