Import any labels you have to start with in whatever format/resolution required. Needs a minimum of 10, ideally 20 records in each class (aka label) to kickstart OL, otherwise manually label. WA has Customer Success consultancy which can generate labels as well as manage a trial.

There are various measures of performance – accuracy, recall, precision etc. The performance is determined against human labelled data as true positive. There’s also ‘folding’ to validate models for stability.

Yes. PrediCX can be used on any text dataset. The rule of thumb is that as long as a third-party human can understand what’s written, then PrediCX should be able to build a predictive model to replicate that human judgement. Clearly the more subjective, or the more technical, the more training will be required and potential uncertainty. The fields used by a human are specified for PrediCX to model.

Yes you can have as many labels as you want per record.

In any business process, things change and things become more important to analyse on a more granular level. PrediCX can easily handle these situations e.g. labels can be merged which updates the models automatically. Also if new labels need to be generated, or new signals appear which require classifying into new labels, then this is all part and parcel of using PrediCX.

PrediCX is built for both data scientists and non-data scientists alike. It is straightforward to use and as long as you know how to label i.e. with domain knowledge around the particular context, then anyone can use it. Having said that, it can massively speed up the work that a data scientist would need to do in generating, validating and deploying machine learning models, so it’s useful to all.

PrediCX’s API makes it straightforward to import data automatically from any system.

The labels and metrics of those labels can be easily exported via the RESTful API to any system.