Fig. 3
From: Visual analytics tool for the interpretation of hidden states in recurrent neural networks
Projection for the Reuters dataset. The data points of the sequences are colored according to their actual class instead of the classification result. In addition, points that belong to correctly classified sequences are more transparent. This visualization helps identify areas where hidden states generate possibly incorrect model outputs. At the top, most sequences were classified as money-fx (blue) instead of crude-oil (green) or grain (red). Underlying data source: Reuters [25]