Hi Ning,

Yes, that could be a problem. You have to set the parameters in a way that this issue does not happen (train_size + test_size well below 1). The usual time series cv also does not test all samples, but I can see your point.

I think MonteCarloCV can be handy for large data sets. If you set train_size to a low value you can still carry many splits without taking too much time.