Artificial Intelligence Now by O'Reilly Media Inc

Author:O'Reilly Media, Inc.
Language: eng
Format: mobi, epub, pdf
Publisher: O'Reilly Media, Inc.
Published: 2017-02-18T05:00:00+00:00

Figure 10-4. Computational graph

The point is that the network definition is simply represented in Python rather than a domain-specific language (DSL), so users can make changes to the network in each iteration (forward computation).

This imperative declaration of neural networks allows users to use standard Python syntax for branching, without studying any DSL. That can be beneficial compared to the symbolic approaches that TensorFlow and Theano utilize and also the text DSL that Caffe and CNTK rely on.

In addition, a standard debugger and profiler can be used to find the bugs, refactor the code, and also tune the hyper-parameters. On the other hand, although Torch and MXNet also allow users to employ imperative modeling of neural networks, they still use the define-and-run approach for building a computational graph object, so debugging requires special care.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.