I want to stop, check and restart an ongoing feedforward neural net.

Suppose i am training a network with net = train(net,x,t) where x inputs, and t targets.

let us say I want to check how well the network is performing at the end of 500 epochs, 1000 epochs etc.

I can manually stop after 500 epochs. and if i am not happy with result i want to further train till 1000 epochs without losing all the previous training. What are the exact steps involved to restart the network from where I left off. I am not adding any thing new by way of data. Everything should remain same but I should be able to continue training further.

Thanks in advance

Answer 1

[Harikrishnan Balachandran Nair]

From my understanding, you want to restart training your network, if it is interrupted in between or you manually stop the process. A possible workaround for this would be to automatically save intermediate checkpoints during the training.

As an example, you can use the following line of code to store the intermediate checkpoints in the file ‘mycheck.mat’ .

net = train(net,x,t,'CheckpointFile','mycheck.mat');

Additionally, you can specify the 'CheckpointDelay' value which decides the frequency at which the checkpoints are stored.