chensaics
(chensaics)
August 4, 2021, 9:53am
1
I am training the Rasa response selector model, but for the epochs, I am not sure what number to set.
I set 100 but the accuracy is 99% at 10 epochs,
I have to stop training and set to 10 and retrain OR waiting for the end of training.
This is time-wasting.
We can set an early stopping function no matter what users set.
This is just a recommendation for improving Rasa,
maybe there is already some functions I do not know?
ChrisRahme
(Chris RahmΓ©)
August 4, 2021, 11:14am
2
Closest thing you can do is set checkpoint_model: True
under DIET, TED, or ResponseSelector, which saves the best performing model during training.
It requires evaluate_on_number_of_examples > 0
and evaluate_every_number_of_epochs > 0
.
1 Like
nik202
(NiK202)
August 4, 2021, 11:50am
3
@chensaics For more details do visit these Github links:
and the one which Chris suggested is this one:
**Proposed changes**:
Implemented model checkpointing for DIET. The best DIET model during training will be stored instead of just the last model. The model is evaluated on the basis of `evaluate_every_number_of_epochs`. Checkpointing is enabled iff the following is set for the `DIETClassifier`, `ResponseSelector` and `TEDPolicy` in the `config.yml` file:
* `checkpoint_model: True`
* `evaluate_on_number_of_examples > 0`
The model is stored to whatever location has been specified with the `--out` parameter when calling `rasa train nlu ...`
**Status (please check what you already did)**:
- [x] added some tests for the functionality
- [x] updated the documentation
- [x] updated the changelog (please check [changelog](https://github.com/RasaHQ/rasa/tree/master/changelog) for instructions)
- [x] reformat files using `black` (please check [Readme](https://github.com/RasaHQ/rasa#code-style) for instructions)
RasaHQ:master
β RasaHQ:model-checkpointing-diet
opened 08:58AM - 10 Jun 20 UTC
Your recommendation are already considered
chensaics
(chensaics)
August 5, 2021, 3:49am
4
@ChrisRahme @nik202 Thank you for your reply!
1 Like
nik202
(NiK202)
August 5, 2021, 7:37am
5
@chensaics No worries close this thread with the solution for others.