Fig. 1
From: Visual analytics tool for the interpretation of hidden states in recurrent neural networks
The main view of our visual analytics approach applied to a model trained on the IMDB dataset. Some general information about the dataset is shown on the panels (A) to (E). (A): The number of input sequences, the classes, and how well the sequences were classified; (B): Visualizations of the different classes to compare the number of input sequences that were correctly classified; (C): This information is also visible in the form of a confusion matrix;Â (D): At the center, the projection of all hidden states produced by the network for all input sequences is visible; (E): On the right side, a list of all sequences allows the selection of a sequence for further exploration. Underlying data source: IMDB as available in Keras [13]