Is there some rule of thumb or method to balance the number of samples vs the sample size. e.g. like a minimum of 10 % sample size or a number of samples = 40% ( like 400 samples for a size of 1000).
My decision was essentially based on the analysis of the confusion matrix. If there are too many unclassified or too many wrong-classified samples, I increased the sample size. You might be also interested on Chapter 5 in my PhD thesis about opinion mining and lexical affect sensing Thesis Opinion Mining and Lexical Affect Sensing
Sample sizes for optimal results is still an open problem. It is mostly based on the choice of modelling techniques, e.g., if you are using parametric methods, the best optimal sample sizes may be 10 to 20% data for training and the remaining for testing via different trials. For non parametric methods, the usual way is about 50% for training and 50% for testing....