3
$\begingroup$

I have trained and validated my LSTM and I would like to deploy it. So, I know that we can save and load the Sequential object of Keras (I am working with Keras as you can guess). I thus implemented a code using these functions.

But, I would like to know whether I must train my model with all the available data (training + test) or just on the training set as I did during my study. Many tutorials talks about how to train a model but not so many are clear about how to deploy a model.

I would like to know what is common to do, knowing that I am doing Time Serie Forecasting which is a specific problem...

Thanks,

$\endgroup$
3
$\begingroup$

Personally, I would side with deploy it as is. I.e. do not retrain on all the data.

Once you train on new data, then you have a new model. You have no idea how this model is going to react to unseen test data coming in from the big wide world. You can’t validate or test it, because you’ve used all that data up in training.

Who knows what might happen!

As you receive more data and get feedback on the model you can create new train/test sets, fine tune the model, then redeploy it as required.

$\endgroup$
  • $\begingroup$ Your answer confirms my feeling. Thank you very much. $\endgroup$ – kakarotto Aug 6 at 13:36
  • $\begingroup$ Glad I could help! I’ve edited the answer slightly to highlight how I would update the model later on. Don’t recreate a new train/test set from scratch, just use new data to help fine tune the model. $\endgroup$ – dijksterhuis Aug 6 at 13:40
  • $\begingroup$ Oh thanks. Now, a quick related question. Can I try to automate this process with callbacks functions. For instance, let's assume that I deployed a given model for a monthly-sampled time serie. Assuming it works after 3 months in the production environment, can I re-train it automatically with a script using callback functions (early stopping, model checkpoints and so on) so that I do not need to work on this as seriously as I did before its deployment ? It would assume a static architecture (layers, dropout) but does it make sense to do so ? Thanks $\endgroup$ – kakarotto Aug 6 at 13:57
  • $\begingroup$ I would strongly advise that you make sure any changes are verified and looked at by a human before it is deployed, as it’s always good to double check it’s not gone haywire. But yes, automation sounds like a sensible idea to reduce your ongoing workload. Checkpoints are good bet to keep a static version for each month, for example. Then you can always rollback to last month if any issues. $\endgroup$ – dijksterhuis Aug 6 at 14:10
  • $\begingroup$ Yes, that makes me aware of the necessity to document correctly what is done so that someone else than me can check and double check things even if some processes are automatized. Thanks again :) hope people will find your answer and upvote it. $\endgroup$ – kakarotto Aug 6 at 15:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.