Animated (3D) t-sne

#1

Hi. I’ m doing the t-sne of a test dataset using this code:

        for batch_ts, (XTest, YTest) in enumerate(test_loader):
            XTest = XTest.to(device)
            recon_batch, mu, logvar, latent_space = model(XTest)
            tsne = TSNE(3, verbose=0)
            tsne_proj = tsne.fit_transform(latent_space.detach().numpy())
            cmap = cm.get_cmap('tab20')
            fig, ax = plt.subplots(figsize=(8, 8))
            num_categories = 7
            for lab in range(num_categories):
                indices = YTest == lab
                ax.scatter(tsne_proj[indices, 0],
                                 tsne_proj[indices, 1],
                                 tsne_proj[indices, 2],
                                 c=np.array(cmap(lab)).reshape(1, 4),
                                 label=lab,
                                 alpha=0.5)
            ax.legend(fontsize='large', markerscale=2)
            plt.show()

I get this error: need at least one array to concatenate. If I do 2D t-sne, the error disappears.
How can I fix it? And is it possible to perform an animated plot of t-sne?

#2

Could you check which method raises the error and check what the expected inputs would be?
I guess a tsne operation might be failing as the inputs might be wrong.