Rasa NLP model training and publishing

[turnerrainer]

AS A Chat Trainer I WANT TO train and publish new NLP models SO THAT its best fit version could be used in production

Acceptance Criteria

• [x] #385
• [x] #386
• [ ] #387

Modifications

• [ ] #420

Technical Stack

• [ ] Store trained Rasa models on Riigipilv S3 File Storage
• [ ] Use Git as a source of truth to save Docker Compose changes
• [ ] Use Cron Manager to initiate recurring tasks
• [x] Deploy all database changes via Liquibase

Scope

• [ ] All Rasa models used are read from an external source
• [x] Train a new model
• [ ] Test the latest model
• [ ] Cross-validate the latest model
• [ ] ~Docker Compose to train new model~
• [ ] ~Docker Compose to test defined model~
• [ ] ~Docker Compose to deploy defined model~

---

Acceptance Criteria

• [x] Train a new NLP model using configuration defined under /training/configuration
• [ ] Deployment of verified models via #387
• [ ] ~Publish a newly trained model (make publish button available if the training is successful)~
• [ ] ~When pressing the "publish" button there will be warning pop up message saying "Are you sure you want to publish? Bot might be unavailable for couple of minutes"~

We need to make a "publish" button for the page /training/train-new-model

DSL

• [x] REST service to train a model
• [ ] ~REST service to publish a model~

[Kristjan259]

Cannot test the scope because it has the some problem as in the task: #635