Estimated time of training and quantity of data required for training

How do you estimate the time of training a bot?
How much training data is required?
What are the parameters that need to be considered while answering these kind of questions?

Hi @KnightCoder,

This depends a lot on your configuration, we don’t have any benchmarks for these questions (would be happy to see them if you experimented yourself, though)

For supervised_embeddings, unless you have >1000 examples in your nlu data then training is normally done in a few seconds. For story training, it’s not so possible to say - depends on the story lengths, use of checkpoints.

You don’t need much training data to get started: 4 or 5 stories, ~50 intent examples.

I understand. But these are usually the questions that come to people while thinking of training a NLP application.

What are all the possible parameters that effect this?

I think once we are sure of the parameters, we might be able to think of at least an approximate formula. Can’t we?