Yes, if I were you I would abandon the idea of swarming and build your own encoder. You want something you can send in a row of input data (in whatever format you want) that returns binary array. The array does not have to be sparse, but be sure to read this paper:
And also you can see examples of existing NuPIC encoders at https://github.com/numenta/nupic/tree/master/src/nupic/encoders. The DateEncoder is a good example of an encoder that joins multiple data input points into one representation.
Once you have an encoder, I would use the anomaly model params for the SP/TM and your own custom tuned encoder params for the new encoder. You might still be able to swarm to get the best params for your new encoder, but since you designed the encoder yourself, you probably have the best idea how to tune it. That’s why I would not suggest swarming if you’re creating your own encoder.