All Implemented Interfaces:
The reset strategy will reset the weights if the neural network fails to fall
below a specified error by a specified number of cycles. This can be useful
to throw out initially "bad/hard" random initializations of the weight
Constructor and Description
int cycles) ResetStrategy
Construct a reset strategy.
Modifier and Type
Method and Description
( init MLTrain train)
Initialize this strategy.
Called just after a training iteration.
Called just before a training iteration.
Methods inherited from class java.lang.
Object clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
public ResetStrategy(double required,
Construct a reset strategy. The error rate must fall
below the required rate in the specified number of cycles,
or the neural network will be reset to random weights and
required - The required error rate.
cycles - The number of cycles to reach that rate.
Copyright © 2014. All Rights Reserved.