General feedback on course

[sbrincat]

Here is some general feedback on the course, compiled from our Miller lab beta testing group. I've already added any typos and minor comments to Anthony's issue. I will compile more general feedback on Neuroblox elsewhere.

• If the course notebooks are to be a primary entry into NB, the models need to actually do what they claim to. Otherwise, the message we are sending to potential new users is that the product does not work, and there is little reason for them to want to continue.
  o The decision making model was not working during the class. I didn’t get a chance to run it again after Haris found that typo, hopefully that fixed it.
  o The corticostriatal model exhibited no behavioral learning. Performance was worse than chance by the end! Is there a bug? Either way, I think the model needs to be tweaked somehow so it actually learns.
  o DCM consistently returned weights that were almost exactly opposite in sign to the true ones, regardless of how we tweaked the model. It was suggested that this was bc DCM performs poorly with little data, but we found it hard to imagine it would be so bad unless there’s some bug. Again, either way, I think the model at least needs to be hand-tweaked (different ground truth? different seed?) so that the fits at least vaguely resemble ground truth.
• Would be good if, in general, the challenge problems were a bit more of an obvious step from the given code—changing specific parameters in the existing model or adding some small well-specified part to it. Rather than a leap to something totally new, eg implementing a totally different model.
• Relatedly, it’s great that the tutorials have links to further reading for users that want to dig deeper into things. But for references needed to understand and complete challenge problems (if we do need them), we should probably try to have those link to very brief summaries rather than multi-page papers or websites.
• All plots should have x/y-axis labels so users know what they are plotting
• In a lot of the text blocks, it looks like the text was intended to be separated into separate lines, but the necessary 2 spaces at EOL were missing, so it just looks like one long run-on line