notebooks aren't great for diffs, but they're great ways to support experimentation.
perhaps want to provide a template repo that installs dinkum, runs any and all notebooks, and saves the results for display.
for big models, how well would it work to put the model in a .model file or something, and then execfile it? Then the notebooks can explore various views of that model, but the model itself is centralized... could provide a separate file containing custom activation functions...
notebooks aren't great for diffs, but they're great ways to support experimentation.
perhaps want to provide a template repo that installs dinkum, runs any and all notebooks, and saves the results for display.
for big models, how well would it work to put the model in a .model file or something, and then execfile it? Then the notebooks can explore various views of that model, but the model itself is centralized... could provide a separate file containing custom activation functions...