Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/715#discussion_r31021740
--- Diff: docs/libs/ml/optimization.md ---
@@ -184,12 +192,15 @@ mini-batch SGD will be performed.
</td>
</tr>
<tr>
- <td><strong>Prediction Function</strong></td>
+ <td><strong>ConvergenceThreshold</strong></td>
<td>
<p>
- Class that provides the prediction function, used to calculate
$\hat{y}$ based on the
- weights and the example features, and the prediction gradient.
- (Default value: <strong>LinearPrediction</strong>)
+ When set, this value acts as a convergence stopping criterion
for the algorithm.
+ If the relative difference in the objective function (loss)
between two consecutive
+ iterations
+ is less than this value we stop the iterations and return the
last weights.
+ This allows us to avoid unnecessary iterations after the
algorithm has converged.
+ (Default value: <strong>None</strong>)
--- End diff --
That would be great IMO
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
