Function generate_plot_lda

plotting/featurespace_visualizations.py:508–559

(feat_path,model_path, target_class,base_class, poison_ids, title, device)

    plt.show()
    
def generate_plot_lda(feat_path,model_path, target_class,base_class, poison_ids, title, device):
    
    [ops_all, labels_all, ids_all] = pickle.load( open( feat_path, "rb" ) ) 
    left_labels = np.array(labels_all)[np.in1d(labels_all,[target_class,base_class])]
    left_ids = np.array(ids_all)[np.in1d(labels_all,[target_class,base_class])]
    left_ops = ops_all[np.in1d(labels_all,[target_class,base_class])]
    id2label = dict(zip(left_ids, left_labels))

    tags = []
    for i in left_ids:
        if i in poison_ids:
            tags.append('poison')
        elif i == 'target':
            tags.append('target')
        elif id2label[i]== target_class:
            tags.append(str(target_class))
        else:
            tags.append(str(base_class))

    tags = np.array(tags)  

    basefeats = left_ops[tags == str(base_class)]
    targfeats = left_ops[tags == str(target_class)]
    poisonfeats = left_ops[tags == 'poison']
    targetfeats = left_ops[tags == 'target']
    basecent = np.mean(basefeats, axis=0)
    targcent = np.mean(targfeats, axis=0)
    poisoncent = np.mean(poisonfeats, axis=0)
   
    ol_tags = np.concatenate([tags[tags == str(base_class)], tags[tags == str(target_class)], tags[tags == str('poison')]])
    ol_feats = np.concatenate([basefeats, targfeats, poisonfeats])
    from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
    lda = LinearDiscriminantAnalysis(n_components=2)
    X_r2 = lda.fit(ol_feats, ol_tags).transform(ol_feats)
    
    colors = ['green', 'royalblue','red']
    target_names = [str(base_class), str(target_class), 'poison']
    alphas = [0.02,0.02,0.5]
    sizes = [5,5,10]
    plt.figure()
    for color, i, target_name,al,si in zip(colors, target_names, target_names,alphas,sizes):
        plt.scatter(X_r2[ol_tags == i, 0], X_r2[ol_tags == i, 1], alpha= al, color=color,
                    label=i,s=si)
    target_proj = lda.fit(ol_feats, ol_tags).transform(targetfeats)
    plt.scatter(target_proj[0][0], target_proj[0][1], alpha=1, color='dimgray', marker='^',label='target',s=50,edgecolors = 'black', linewidth=3)
    plt.xlabel('LD1',fontsize=15,fontweight='medium',fontvariant='small-caps')
    plt.ylabel('LD2',fontsize=15,fontweight='medium',fontvariant='small-caps')
    figname = title.replace(" ", "_")+ "_3d_lda.pdf"
    plt.savefig(os.path.join('./plots', figname), bbox_inches='tight')
    #     plt.legend(loc='best', shadow=False, scatterpoints=1)
#     plt.title(title)
    plt.show()

def genplot_centroid_prob_3d(feat_path, model_path, target_class,base_class, poison_ids, title, device):