You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@singa.apache.org by "wangwei (JIRA)" <ji...@apache.org> on 2016/05/20 11:58:12 UTC
[jira] [Created] (SINGA-176) Add loss and metric base classes
wangwei created SINGA-176:
-----------------------------
Summary: Add loss and metric base classes
Key: SINGA-176
URL: https://issues.apache.org/jira/browse/SINGA-176
Project: Singa
Issue Type: New Feature
Reporter: wangwei
The loss base class 'Loss' is for learning objectives, which accept the prediction from the neural net and the target (or ground truth) from the training dataset. It outputs a scalar loss value for each data instance and computes the gradient of the loss value w.r.t the prediction value, which would be back-propagated through the neural net.
The metric base class 'Metric' is for evaluating the performance (e.g, accuracy) of the neural net. It also accepts the prediction and the target, and computes the performance metrics, which could be accuracy, false positive, etc. It does not compute the gradients.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)