Rasa spliting algorithm does not give precise number of training samples.

[duongkstn]

Description of Problem: Some examples may be missed due to Rasa spliting algorithm.The issues is clearly depicted at forum thread

Overview of the Solution: rasa data split does not give precise number of training samples.
Say overall we have X samples (x₁ samples of label l₁, x₂ samples of label l₂, …) and training-fraction is 0.8.
(Note: x₁ + x₂ + … = X).
In the code of Rasa , number of training samples is A = int(0.8 * x₁) + int(0.8 * x₂) + …
Mathematically, A ≤ int(0.8 * X).
So number of missing samples is int(0.8 * X) - A

[sara-tagger]

Thanks for submitting this feature request 🚀 @JustinaPetr will get back to you about it soon! ✨

[tabergma]

closed via #7095