Other Tools

TensorBoard

• Web-based software used for visualizing the training of the model.
• The example above shows the training of the Seq2Seq model with the blue line representing the training set and the orange line representing the evaluation set.
• A downward trend in the graphs should be observed when training. An increasing lost value of the evaluation set (orange line) shows the model being overfitted.
• Generally useful for monitoring the training of models.

Running the Program

1. Open the terminal and enter the following command.

   python -m tensorflow.tensorboard --logdir <enter log directory> --port <port number>

Attention Visualizer

• Displays the source text and generated summary.
• Shows the heatmap of attention for each generated word. (Yellow words in the source text)
• Shows the Pgen values for Pointer-Generator models. (Green words in the generated summary)

Running the Program

1. Download the code from the Attention Visualizer Github

2. Place the external folder and index.html into the decode directory

3. Open the terminal, cd into the decode directory, and enter the following command.

   python -m SimpleHTTPServer

4. Navigate to the browser on

   http://<ip-address>:8000/

Other Resources & Dependencies

• TensorFlow TensorBoard Documentation
• Attention Visualizer Github