[metalearn] run experiment with fewer number of tasks

[cosmicBboy]

To validate whether or not the RNN weights of the controller implements an RL algorithm, train and test with a smaller set of datasets.