online learning for label encoder and random forest classifier

Question

I have a very large dataset that needs to be used for classification, I sampled the data, but that does not guarantee that I will have the whole labels in my output. How can I sample my data to cover all labels? Also, I wanted to save the label encoder and the RandomForestClassifier that I used in this process to use them for incremental learning. I searched about using the RandomForestClassifier I found out that set_warm feature is only used for adding extra estimators not updating the weights. and partial_fit does not support using random forest. So my second question is how to updated the label encoder and the RandomForestClassifier for training another datasets that might have other labels and more datapoints?