February 2018
Intermediate to advanced
378 pages
10h 14m
English
When our usual code has a bug, it either doesn't work or works in the wrong way. When ML code has a bug, it often continues working but just degrades in quality. Because machine learning algorithms can be extremely complex, good debugging and visualization tools are of extreme value. For TensorFlow for example, such a tool is TensorBoard, which allows exploring model graphs, weight distributions, loss charts, and so on.
For now, humanity has not invented a better way to understand data than to visualize it. Often, 10 minutes of writing code for visualization lead to more insights than hours of debugging on a console. As Prof. Ben Shneiderman from the University of Maryland once noted in his talk:
Read now
Unlock full access