Best practice of managing original data once model is trained and retrained again

[tankhar]

Hi,
My team needs to deal with a scenario where we will get continuous stream of data points.(training data)
We think that once we get a specific number of samples, we will train an ML.net Model from these.
Now, these data points will keep coming , so we may need retrain the model again to accommodate the incremental data points
Questions regarding this:

1. After we train the ML.Net model first time, is it ok to discard the original dataset that we had?(we will save on space and since data will keep coming contionously we cannot keep it forever)
2. When we retrain the ML.Net model(incremental learning), is there a risk of losing accuracy?
   3 Is there a better way to do this?
3. What are the best practices of managing orginal data whether we do batch learning or incremental learning?

Thanks