Function plot_pair_distribution

utils.py:185–216

(type, X, clean_index, noisy_index, save_path='')

                    for param in net.parameters():
                        param.requires_grad = requires_grad
def plot_pair_distribution(type, X, clean_index, noisy_index, save_path=''):
    plt.clf()
    ax = plt.gca()

    font1 = {'family': 'Times New Roman',
             'weight': 'normal',
             'size': 13,
             }

    # Plot data histogram
    if type == 0:
        ax.hist(X[clean_index], bins=100, density=True, histtype='stepfilled', color='darkorchid', alpha=0.3,
                label='True Pos. Pairs')
        ax.hist(X[noisy_index], bins=100, density=True, histtype='stepfilled', color='blue', alpha=0.3,
                label='False Pos. Pairs')
    if type == 1:
        ax.hist(X[clean_index], bins=100, density=True, histtype='stepfilled', color='teal', alpha=0.3,
                label='True Neg. Pairs')
        ax.hist(X[noisy_index], bins=100, density=True, histtype='stepfilled', color='peru', alpha=0.3,
                label='False Neg. Pairs')

    ax.set_xlabel('Normalized Distances', fontdict=font1)
    ax.set_ylabel('Frequency', fontdict=font1)
    x_ticks = np.array([0.0, 0.2, 0.4, 0.6, 0.8, 1.0])
    plt.xticks(x_ticks)
    plt.tick_params(labelsize=11)
    ax.legend(loc='upper left', prop=font1)

    if save_path:
        plt.savefig(save_path, dpi=300)
    else:
        plt.show()
                        
class AllSampler(Sampler):
    def __init__(self, dataset, train_color_label, train_thermal_label, shuffle=True):