Problem using unsupervised helloworld example ( Similarity results on CIFAR10 dataset )
robertorovella91 opened this issue · comments
Roberto commented
Hi everyone, I tried the unsupervised hello world example from the notebook, but after training, despite the classifier results leading to a significant improvement in terms of accuracy, when I go to test the algorithm in terms of similarity, the results reported by the knn search seem not to work (using the same example of the supervised helloworld notebook).
I also tried to run the model's prediction and calculate the cosine similarity between different images but it returns always 1. What am I doing wrong?