def callback_recovery(loc): d = loc["dict_obj"] d.wc.append(emd(loc["dictionary"], d.generating_dict, "chordal", scale=True)) d.wfs.append(emd(loc["dictionary"], d.generating_dict, "fubinistudy", scale=True)) d.wcpa.append( emd(loc["dictionary"], d.generating_dict, "chordal_principal_angles", scale=True) ) d.wbc.append(emd(loc["dictionary"], d.generating_dict, "binetcauchy", scale=True)) d.wg.append(emd(loc["dictionary"], d.generating_dict, "geodesic", scale=True)) d.wfb.append(emd(loc["dictionary"], d.generating_dict, "frobenius", scale=True)) d.hc.append(hausdorff(loc["dictionary"], d.generating_dict, "chordal", scale=True)) d.hfs.append( hausdorff(loc["dictionary"], d.generating_dict, "fubinistudy", scale=True) ) d.hcpa.append( hausdorff( loc["dictionary"], d.generating_dict, "chordal_principal_angles", scale=True ) ) d.hbc.append( hausdorff(loc["dictionary"], d.generating_dict, "binetcauchy", scale=True) ) d.hg.append(hausdorff(loc["dictionary"], d.generating_dict, "geodesic", scale=True)) d.hfb.append(hausdorff(loc["dictionary"], d.generating_dict, "frobenius", scale=True)) d.dr99.append(detection_rate(loc["dictionary"], d.generating_dict, 0.99)) d.dr97.append(detection_rate(loc["dictionary"], d.generating_dict, 0.97))
def callback_recovery(loc): d = loc['dict_obj'] d.wc.append(emd(loc['dictionary'], d.generating_dict, 'chordal', scale=True)) d.wfs.append(emd(loc['dictionary'], d.generating_dict, 'fubinistudy', scale=True)) d.hc.append(hausdorff(loc['dictionary'], d.generating_dict, 'chordal', scale=True)) d.hfs.append(hausdorff(loc['dictionary'], d.generating_dict, 'fubinistudy', scale=True)) d.bd.append(beta_dist(d.generating_dict, loc['dictionary'])) d.dr99.append(detection_rate(loc['dictionary'], d.generating_dict, 0.99)) d.dr97.append(detection_rate(loc['dictionary'], d.generating_dict, 0.97))
def callback_recovery(loc): d = loc["dict_obj"] d.wc.append( emd(loc["dictionary"], d.generating_dict, "chordal", scale=True)) d.wfs.append( emd(loc["dictionary"], d.generating_dict, "fubinistudy", scale=True)) d.hc.append( hausdorff(loc["dictionary"], d.generating_dict, "chordal", scale=True)) d.hfs.append( hausdorff(loc["dictionary"], d.generating_dict, "fubinistudy", scale=True)) d.bd.append(beta_dist(d.generating_dict, loc["dictionary"])) d.dr99.append(detection_rate(loc["dictionary"], d.generating_dict, 0.99)) d.dr97.append(detection_rate(loc["dictionary"], d.generating_dict, 0.97))
def test_correlation(): du2 = [randn(n_features,) for i in range(n_kernels)] for i in range(len(du2)): du2[i] /= norm(du2[i]) dm2 = [randn(n_features, n_dims) for i in range(n_kernels)] for i in range(len(dm2)): dm2[i] /= norm(dm2[i]) assert_equal(100., detection_rate(du, du, 0.97)) assert_not_equal(100., detection_rate(du, du2, 0.99)) assert_equal(100., detection_rate(dm, dm, 0.97)) assert_not_equal(100., detection_rate(dm, dm2, 0.99)) assert_equal((100., 100.), precision_recall(du, du, 0.97)) assert_equal((0., 0.), precision_recall(du, du2, 0.99)) assert_true(allclose(precision_recall_points(du, du), (ones(len(du)), ones(len(du))))) assert_true(not allclose(precision_recall_points(du, du2), (ones(len(du)), ones(len(du2)))))
def test_correlation(): du2 = [randn(n_features, ) for i in range(n_kernels)] for i in range(len(du2)): du2[i] /= norm(du2[i]) dm2 = [randn(n_features, n_dims) for i in range(n_kernels)] for i in range(len(dm2)): dm2[i] /= norm(dm2[i]) assert_equal(100., detection_rate(du, du, 0.97)) assert_not_equal(100., detection_rate(du, du2, 0.99)) assert_equal(100., detection_rate(dm, dm, 0.97)) assert_not_equal(100., detection_rate(dm, dm2, 0.99)) assert_equal((100., 100.), precision_recall(du, du, 0.97)) assert_equal((0., 0.), precision_recall(du, du2, 0.99)) assert_true( allclose(precision_recall_points(du, du), (ones(len(du)), ones(len(du))))) assert_true(not allclose(precision_recall_points(du, du2), (ones(len(du)), ones(len(du2)))))
def callback_distance(loc): ii, iter_offset = loc['ii'], loc['iter_offset'] n_batches = loc['n_batches'] if np.mod((ii-iter_offset)/int(n_batches), n_iter) == 0: # Compute distance only every 5 iterations, as in previous case d = loc['dict_obj'] d.wasserstein.append(emd(loc['dictionary'], d.generating_dict, 'chordal', scale=True)) d.detect_rate.append(detection_rate(loc['dictionary'], d.generating_dict, 0.99)) d.objective_error.append(loc['current_cost'])
def callback_distance(loc): ii, iter_offset = loc["ii"], loc["iter_offset"] n_batches = loc["n_batches"] if np.mod((ii - iter_offset) / int(n_batches), n_iter) == 0: # Compute distance only every 5 iterations, as in previous case d = loc["dict_obj"] d.wasserstein.append( emd(loc["dictionary"], d.generating_dict, "chordal", scale=True)) d.detect_rate.append( detection_rate(loc["dictionary"], d.generating_dict, 0.99)) d.objective_error.append(loc["current_cost"])
# dict_init[i] /= norm(dict_init[i], 'fro') dict_init = None learned_dict = MiniBatchMultivariateDictLearning(n_kernels=n_kernels, batch_size=batch_size, n_iter=n_iter, n_nonzero_coefs=n_nonzero_coefs, n_jobs=n_jobs, learning_rate=learning_rate, kernel_init_len=kernel_init_len, verbose=1, dict_init=dict_init, random_state=rng_global) # Update learned dictionary at each iteration and compute a distance # with the generating dictionary for i in range(max_iter): learned_dict = learned_dict.partial_fit(X) # Compute the detection rate detect_rate.append(detection_rate(learned_dict.kernels_, generating_dict, 0.99)) # Compute the Wasserstein distance wasserstein.append(emd(learned_dict.kernels_, generating_dict, 'chordal', scale=True)) # Get the objective error objective_error.append(learned_dict.error_.sum()) plot_univariate(array(objective_error), array(detect_rate), array(wasserstein), n_iter, 'univariate-case') # Another possibility is to rely on a callback function such as def callback_distance(loc): ii, iter_offset = loc['ii'], loc['iter_offset'] n_batches = loc['n_batches'] if np.mod((ii-iter_offset)/int(n_batches), n_iter) == 0: # Compute distance only every 5 iterations, as in previous case