With Re:infer's comprehensive Validation capabilities, including the Model Rating functionality, understanding when to stop training your model is now relatively simple.
The required level of performance for your model will be up to your and your business, but Re:infer's Model Rating gives you a great sense of where your model is performance-wise, and how to improve it if needed.
A model with a score of 70 or above is classed by Re:infer as 'Good', whilst a score of 90 is required for a model to be classed as 'Excellent'.
Whatever the use case, we would always recommend ensuring the following before stopping training:
- That your model at least has an overall score that provides a rating of 'Good' as this means Re:infer considers the model to relatively healthy overall
- That each of the individual factors also have a rating of at least 'Good' (as shown below)
- That none of your important labels have red or amber performance warnings
For an analytics focused model, beyond the factors listed above it should be at the model trainer's discretion how much they want to optimise their model's performance. The performance requirement can depend on a variety of factors, including the objectives of the use case and the capacity of the model trainer to continue training.
If you're creating a model that is intended to enable automations, it's recommended that your model should have an 'Excellent' rating, and also that the model is tested on live data before being deployed into production.
Example model rating for an extremely health model
Additional optional performance checks
Whilst the Model Rating is a comprehensive performance assessment, you may want to complete some additional checks to ensure you're completely comfortable with the performance of your model.
If this is the case, here are some useful checks you can do with recommended actions. It's worth noting that if Re:infer thinks it's important for you to take any of these actions, it will also be recommending these in Validation.
| Check | Process | Actions to take |
|---|---|---|
| 2-day period prediction review | Review predictions on 1-2 days worth of recent data: use the time filter and ‘recent’ in the drop down to pick 2 recent days worth of data. Review the predictions and make sure each verbatim has a reasonably high confidence prediction. By reviewing predictions 1-2 days worth of data it should ensure all potential concepts are covered | • If there are verbatims with no predictions or insufficient confidence then label them as normal • Then train more in Shuffle and Low Confidence |
| Shuffle | Review predictions in Shuffle for at least 5 pages. Each verbatim should have a label predicted with a reasonably high confidence | • If there are verbatims with no predictions or insufficient confidence then label them as normal • Then train more in Shuffle and Low Confidence |
| Low Confidence | Low Confidence mode shows you verbatims where the confidence of any label applying is low. This should be used to ensure you have not missed any potential labels or that you have provided enough examples for each existing label | • If there are verbatims that have not been covered add a new label for them and train out as normal • Where you find a verbatim for an existing label, apply it as normal |
| 'Re-Discover' (see below) | Returning to Discover can show you potential new clusters where the probability of any label applying is low. This should be used to ensure you have not missed any potential labels or to provide existing labels with more varied examples, in a similar way to Low Confidence | • If there are clusters with no predictions (or very low) then label the cluster with either a new label or an existing one if applicable • Train out any new label as normal |
'Re-Discover'
'Re-Discover' is a step that can be revisited at any time during the training process, but can also be useful when checking whether you have completed sufficient training.
This check essentially just entails going back to the Discover page on 'Cluster' mode and reviewing the clusters in there to check their predictions and to see if Discover has found any clusters that may have been missed by your training.
As the clusters in Discover retrain after a significant amount of training has been completed in the platform (180 annotations) or a significant amount of data has been added to the dataset (1000 verbatims or 1%, whichever is greater, and at least 1 annotation), they should regularly update throughout the training process.
Discover tries to find clusters that are not well covered by label predictions. If there are clusters in Discover that should have certain labels predicted but don't, you know you need to do some more training for those labels. See here for how to label clusters in Discover.
If your model is well trained, Discover will struggle to find clusters with low confidence or no predictions. If you see that each of the clusters in Discover has reasonably high-confidence and correct predictions, this is a good indicator that your model covers the dataset well.
Previous: Improving Balance and using 'Rebalance'