Computer Vision News - June 2019

26 Computer Vision News Focus on It is also possible to use some cooler visualizations using the same embedding. For example, we can write the digit of each cluster instead of each scatter. This is performed by replacing the for loop in our plotting function with this piece of code: This new visualization gives us the following image: Lastly, we can also attach an image to each of the clusters to get even more appealing results. This gives us our final results, displayed on the next page. Conclusion As we have seen, t-SNE can be a very useful tool to understand high dimensional data . A widely used application of t-SNE is to visualize, in the way we did, the last layer of a neural network. In such visualization, one can see the level of separability of the last layer. Of course, it needs to be linearly separable since the last layer is only linear. We Tried for You for i in range ( X . shape [ 0 ]): plt . text ( X [ i , 0 ], X [ i , 1 ], str ( y [ i ]), color = plt . cm . Set1 ( y [ i ] / 10. ), fontdict = { 'weight' : 'bold' , 'size' : 9 } )