Pass SDRs to a network model w/o FileRecordStream


Is there a way to just send already encoded SDRs through a PassThroughEncoder w/o using FileRecordStream to a network? I.e. “not specify” a dataSource.

I manage to do this w/o specifying a network when only using a TM object. But thought it would be nice to use a network structure.

To recap my intended algorithm:

i. read next line in a pandas df (i.e. a sequence)
ii. encode each item to via a custom SDR
iii. pass each SDR to the network via the PassThroughEncoder
iv. reset TM and SP after each sequence



I am currently doing the same by using the PluggableEncoderSensor with an encoder of your choice and changing the sensed value on each Iteration before calling run(1).
You could define a trivial encoder there or encode directly in the network process and send in the pandas df.

Maybe not an elegant solution, but it would full fill your purpose.

Kind regards

Thanks! It was probably the yaml.dump of parameters into the PassThroughEncoder that wasn’t working for me and got me worried. But everything is working now. :smiley:

If anyone else ever needs some tips this was a good example: