Don't know where your data is from? Bayesian modeling for unknown coordinates

Summary

location_surface_grids = {} with location_error_gp_model: for multiplier, X_noisy_value in zip(multipliers, noisy_xs): pm.set_data({"X_noisy": X_noisy_value, "σ_s": multiplier}) posterior_mean_point = { name: location_error_idatas[multiplier].posterior[name].mean(("chain", "draw")).values for name in ["μ", "σ", "ℓ", "σ0", "Δs", "X_true"] } f_mean, _ = gp_location.predict(Xnew, point=posterior_mean_point, diag=True, pred_noise=False) location_surface_grids[multiplier] = f_mean.reshape(n_prediction_grid, n_prediction_grid) naive_kde_grids = {} kde_bandwidths = { multiplier: location_error_idatas[multiplier].posterior["ℓ"].mean(("chain", "draw")).item() for multiplier in multipliers } for multiplier, X_noisy_value in zip(multipliers, noisy_xs): squared_distance = ( (Xnew[:, 0, None] - X_noisy_value[:, 0]) ** 2 + (Xnew[:, 1, None] - X_noisy_value[:, 1]) ** 2 ) weights = np.exp(-0.5 * squared_distance / kde_bandwidths[multiplier]**2) naive_kde_grids[multiplier] = (weights @ y_walker / weights.sum(axis=1)).reshape(n_prediction_grid, n_prediction_grid) location_norm = colors.Normalize( vmin=min(y_walker.min(), *(grid.min() for grid in location_surface_grids.values())), vmax=max(y_walker.max(), *(grid.max() for grid in location_surface_grids.values())), ) point_colors = [theme["gunmetal"], theme["sepia"], theme["rust"], theme["steel"]] arrow_head_width = 0.018 * max(x_range, y_range) fig = plt.figure(figsize=(5, 11.2 * (5 / 7))) gs = fig.add_gridspec(4, len(multipliers), hspace=0.08, wspace=0.08) axes = np.array([[fig.add_subplot(gs[row, col]) for col in range(len(multipliers))] for row in range(4)]) panel_idx = 0 surface_meshes = [] for col, (multiplier, X_noisy_value) in enumerate(zip(multipliers, noisy_xs)): ax = axes[0, col] ax.scatter(X_walker[:, 0], X_walker[:, 1], facecolor=theme["paper"], edgecolor=theme["gunmetal"], s=14, linewidth=0.35, alpha=0.5) ax.scatter(X_noisy_value[:, 0], X_noisy_value[:, 1], c=y_walker, cmap=sepia_cmap, norm=location_norm, s=16, edgecolor=...